Computer Modelling of

Electrical Power Systems

 Computer Modelling of
Electrical Power Systems
J. ARRILLAGA and C. P. ARNOLD
Department of Electrical Engineering,
University of Canterbury, New Zealand

and

B. J . HARKER
New Zealand Electricity

A Wiley-Interscience Publication

JOHN WILEY & SONS

Chichester - New York . Brisbane Toronto . Singapore
 Copyright O 1983 by John Wiley & Sons Ltd.

Reprinted April 1991

All rights reserved.

No part of this book may be reproduced by any means, nor

transmitted, nor translated into a machine language
without the written permission of the publisher.

Library of Congress Cataloging in Publication Data:

Arrillaga, J.
Computer modelling of electrical power systems.
'A Wiley-Interscience publication.'
Includes bibliographies and index.
.l. Electric power systems-Mathematical models.
.' 2. Electric power systems-Data processing.
I. Arnold, C. P. 11. Harker, B. J. 111. Title.
TK1005.A76 621.319' 1'0724 82-2664

ISBN 0 471 10406 X AACR2

#* <'

British Library Cataloguing in Publication Data:

Arrillaga, J .
Computer modelling of electrical power systems.
1. Electric power systems-Simulation methods
2. Electric power systems-Data processing
I. Title 11. Arnold, C. P. 111. Harker, B. J.
621.31'0724 TKlOO5

Photosetting by Thomson Press (India) Ltd., New Dehli

Printed by Antony Rowe Ltd., Chippenham, Wiltshire
 Contents
...
PREFACE . . Xlll

1. INTRODUCTION . 1
1.1. General background . . 1
1.2. Transmission system development . 2
1.3. Theoretical models and computer programs . 3

2. MODELLING O F POWER TRANSMISSION PLANT .

2.1. Introduction .
2.2. Linear transformation techniques. .
2.3. Basic single-phase modelling .
Transmission lines; Transformer on nominal ratio; Off-
nominal transformer tap settings; Phase-shifting trans-
formers
2.4. Three-phase system analysis .
Discussion of the frame of reference; The use of compound
admittances; Rules for forming the admittance matrix of
simple networks; Network subdivision
2.5. Three-phase models of transmission lines .
Mutually coupled three-phase lines; Consideration of ter-
minal connections; Shunt elements; Series elements
2.6. Three-phase models of transformers .
Primitive admittance model of three-phase transformers;
Models for common transformer connections; Three-
phase transformer models with independent phase tap
control; Sequence components modelling of three-phase
transformers; Line sectionalization
2.7. Formation of the system admittance matrix .
2.8. References .

3. MODELLING O F STATIC A. C.-D. C.

CONVERSION PLANT . . 43
3.1. Introduction . . 43
3.2. Rectification . . 44
 3.3. Inversion . . 48
3.4. Commutation reactance . . 50
3.5. D.C. transmission . . 53
Alternative forms of control
3.6. References . . 57

4. LOAD FLOW . 58
4.1. Introduction . 58
4.2. Basic nodal method . . 59
4.3. Conditioning of Y matrix . 61
4.4. The case where one voltage is known . 62
4.5. Analytical definition of the problem . 64
4.6. Newton-Raphson method of solving load flows . . 66
Equations relating to power-system load flow
4.7. Techniques which make the Newton-Raphson
method competitive in load flow . 72
Sparsity programming; Triangular factorization; Optimal
ordering; Aids to convergence
4.8. Characteristics of the Newton-Raphson
load flow . . 75
4.9. Decoupled Newton load flow . 76
4.10. Fast decoupled load flow . . 78
4.11. Convergence criteria and tests . 82
4.12. Numerical example . 83
4.13. References . . 88

5. THREE-PHASE LOAD FLOW . 90

5.1. Introduction . . 90
5.2. Notation . . 91
5.3. Formulation of the three-phase load-flow problem . . 91
Synchronous machine modelling; Specified variables;
Derivation of equations
5.4. Fast decoupled threephase algorithm . . 96
Jacobian approximations; Generator models and the fast
decoupled algorithm
5.5. Structure of the computer program . . 103
Data input; Factorization of constant Jacobians; Starting
values; Iterative solution; Output results
5.6. Performance of the algorithm . . 105
Performance under balanced conditions; Performance with
unbalanced systems
 vii
5.7. Test system and results .
Test cases and typical results
5.8. References .
6. A. C.-D. C. LOAD FLOW .
6.1. Introduction .
6.2. Formulation of the problem. .
6.3. D. C. system model .
Converter variables; D. C. per unit system; Derivation of
equations; Incorporation of control equations; Inverter
operation
6.4. Solution techniques .
Unified solution; Programming considerations for the
unified algorithms; Sequential method
6.5. Control of converter a.c. terminal voltage .
6.6. Extension to multiple and/or multiterminal
d.c. systems. .
6.7. D. C. convergence tolerance .
6.8. Test system and results .
Unified cases; Sequential cases; Initial conditions for d.c.
system; Effect of a.c. system strength; Discussion of con-
vergence properties
6.9. Numerical example .
6.10. References .
7. THREE-PHASE A. C.-D. C. LOAD FLOW .
7.1. Introduction .
7.2. Formulation of the three-phase a.c.-d.c.
load-flow problem .
7.3. D. C.-system modelling .
Basic assumptions; Selection of converter variables;
Converter angle reference; Per unit system; Converter
source voltages; D. C. voltage; D. C. interconnection; In-
corporation of control strategies; Inverter operation with
minimum extinction angle; Enlarged converter model;
Remaining twelve equations; Summary of equations and
variables
7.4. Load-flow solution .
7.5. Program structure and computational aspects .
D. C. input data; Programming aspects of the iterative
solution
7.6. Performance of the algorithm .
Test system; Convergence of d.c. model from fixed terminal
 ...
Vlll

conditions; Performance of the integrated a.c.-d.c. load

flow; Sample results; Conclusions on performance of the
algorithm
Effect of harmonic voltages on the fundamental
frequency operating state of d.c. converters .
Harmonic voltage limits; Shift in zero crossing of phase-
phase voltages; Effect of crossing error on the average d . ~ .
voltage; Errors in calculated phase current magnitudes
References .
8. FAULTED SYSTEM STUDIES .
8.1. Introduction .
8.2. Analysis of three-phase faults .
Admittance matrix equation; Impedance matrix equation;
Fault calculations
8.3. Analysis of unbalanced faults .
Admittance matrices; Fault calculations; Short-circuit
faults; Open-circuit faults
8.4. Program description and typical solutions .
8.5. References .

9. POWER SYSTEM STABILITY-BASIC MODEL .

9.1. Introduction .
The form of the equations; Frames of reference
9.2. Synchronous machines- basic models . .
Mechanical equations; Electrical equations
9.3. Synchronous machine automatic controllers . .
Automatic voltage regulators; Speed governors; Hydro and
thermal turbines; Modelling lead-lag circuits
9.4. Loads. .
Low-voltage problems
9.5. The transmission network . . .
9.6. Overall system representation .
Mesh matrix method; Nodal matrix method; Synchronous
machine representation in the network; Load represen-
tation in the network; System faults and switching
9.7. Integration . .
Problems with the trapezoidal method; Programming the
trapezoidal method; Application of the trapezoidal method
9.8. Structure of a transient stability program .
Overall structure; Structure of machine and network
iterative solution
9.9. General conclusions .
9.10. References .
10. POWERSYSTEM STABILITY-ADVANCED
COMPONENT MODELLING .
10.1. Introduction .
10.2. Synchronous machine saturation .
Classical saturation model; Salient machine saturation;
Simple saturation representation; Saturation curve repre-
sentation; Potier reactance; The effect of saturation on the
synchronous machine model; Representation of saturated
synchronous machines in the network; Inclusion of synch-
ronous machine saturation in the transient stability
program
10.3. Detailed turbine model .
10.4. Induction machines .
Mechanical equations; Electrical equations; Electrical
equations when the slip is large; Representation of in-
duction machines in the network; Inclusion of induction
machine in the transient stability program
10.5. A. C.-D. C. conversion .
Rectifier loads; D. C. link; Representation of converters in
the network; Inclusion of converters in the transient
stability program
10.6. Static var compensation systems .
Representation of SVS in the overall system
10.7. Relays .
Instantaneous overcurrent relays; Inverse definite mi-
nimum time lag overcurrent relays; Undervoltage relays;
Induction machine contactors; Directional overcurrent
relay; Distance relays; Incorporating relays in the transient
stability program
10.8. Unbalanced faults .
Negative-sequence system; Zero-sequence system; In-
clusion of negative and zero-sequence systems for un-
symmetrical faults
10.9. General conclusions .
10.10. References .
11. DYNAMIC SIMULATION OF A. C.-D. C. SYSTEMS .
11.I. Introduction .
11.2. Formulation of the differential equations .
Definitions; Voltage and current relations; Evaluation of
dependent variables; Formulation for variable topology
11.3. Transient converter model .
Modes of operation; Converter topology; &nodes and
basic L,; Representation of modes of operation; Valve
 voltages and currents; Initial conditions; Converter control
system
11.4. Transient generator model .
11,5. Transmission lines .
11,6. Transient transformer model .
Transformer saturation
11.7. Static elements .
11.8. A. C. system .
Time variant Thevenin equivalents
11,9. D. C. system .
11.10.Solutionofthenetworkequations .
Numerical integration; Choice of state variables; Per unit
system; Solution of the transformed network equations;
Discontinuities
11.11. Computer implementation .
Input and establishing network equations; Monitoring and
control of converter bridges; External disturbances;
Output; Interactive coordination with transient stability
analysis; Incorporation of time variant network equivalents
11.12. Program optimization and use .
Scope; Sparsity storage; Coefficient matrices; Incidence
matrices; Implementation
11.13. General conclusions .
11.14. References .

-D. C. SYSTEM DISTURBANCES .

Introduction .
A. C. FAULT MODELLING
Reduced network representation .
Fault application and removal .
Equivalent a.c. system representation; Full a.c. system
representation
Test system and results .
D. C. FAULT MODELLING
Overcurrent characteristics of d.c. line faults .
Detection of d.c. line faults.
Fault characteristics .
Arc extinction and deionization; Converter restart
Results .
Fault application; Fault detection; Arc extinction and
deionization; D.c. line recharge and power transmission
restart; Full simulation of a d.c. line fault
References .
13. TRANSIENT STABILITY ANALYSIS O F
A.C.-D.C.SYSTEMS. .
13.1. Introduction .
13.2. Requirements of the transient stability program .
13.3. Obtaining information from the
TCS program . . .
Spectral analysis; Simplification in obtaining results
13.4. Incorporating transient converter simulation
in a transient stability study .
Solution at the converter node using TCS input; Alignment
of TCS input with TS
13.5. Modification of transient stability
program for a.c. faults .
Algorithm for simultaneous fault and link solution
13.6. Results using the coordinated program. .
Rectifier a.c. system fault; Inverter a.c. system fault; D.c.
line fault; Conclusion
13.7. Overall conclusions .
13.8. References .
APPENDICES

I. NUMERICAL INTEGRATION METHODS .

1.1. Introduction .
1.2. Properties of the integration methods .
Accuracy; Stability; Stiffness
1.3. Predictor-corrector methods .
1.4. Runge-Kutta methods. .
1.5. References .
11. POWERSYSTEM DATA .
SUBJECT INDEX . . 420
 Preface
This book describes the use of power-system component models and efficient
computational techniques in the development of a new generation of computer
program representing the steady and dynamic states of Electrical Power Systems.
The content, although directed to practicing engineers, should also be
appropriate for Advanced Power System courses at final year degree and Masters
levels. Much of the material included was developed by the authors and their
colleagues at the University of Manchester Institute of Science and Technology
and used as part of a postgraduate course in Electrical Power Systems Analysis
and Control. The integration of h.v.d.c. transmission and the final development of
programs were carried out by the authors since 1975 at the University of
Canterbury (New Zealand), in collaboration with New Zealand Electricity.
Some basic knowledge of Power-System theory, matrix analysis and numerical
techniques is presumed, and specific references are given to help the uninitiated to
pick up the relevant material.
An introductory chapter describes the main computational and transmission
system developmentsjustifying the purpose of the book. Steady state models of
a.c. and d.c. power-system plant components with varying degrees of complexity
are derived in Chapters 2 and 3 respectively.
A general purpose single-phase a.c. load-flow program is described in Chapter
4 with particular reference to the Newton Fast-Decoupled algorithm. In Chapter
5, the decoupling technique is used to develop a three-phase load-flow program
and Chapters 6 and 7 extend the previous algorithms to incorporate static
converters and h.v.d.c. transmission. Chapter 8 is devoted to the modelling of
conventional a.c. faults.
The remaining chapters consider the power system in the dynamic state.
Chapter 9 describes basic dynamic models of a.c. power-system plant and their
use in multi-machinetransient stability studies. More advanced dynamic models
and a quasi-steady-state representation of large converter plant and h.v.d.c.
transmission are developed in Chapter 10.
A detailed Transient Converter Simulation model and its application to the
analysis of a.c. and d.c. short-circuit faults are described in Chapters 11 and 12
respectively.
Finally, Chapter 13 combines the Transient Converter Simulation and multi-
...
Xlll
xiv
machine transient stability programs for the assessment of transient stability in
a.c. systems containing h.v.d.c. transmission links.
The authors should like to express their gratitude to P.W. Blakeley (ex-
Manager) and K. McCool present General Manager of New Zealand Electricity
for their continuous support in this work. They also wish to acknowledge the
considerable help received over a decade from many of their colleagues and in
particular from A. Brameller, B. Stott, M. D. Heffernan, K. S. Turner, P. Bodger,
J. G. Campos Barros, H. Al-Khashali and D. B. Giesner. Finally, the authors are
very grateful to Mrs. A. Haughan for her active cooperation in the preparation of
the manuscript.
 Introduction
1.1. GENERAL BACKGROUND
In parallel with the development of larger and faster digital computers in the
1950's, considerable effort was devoted to the computer modelling of large
Electrical Power Systems. Several good books have collected such experience
emphasizing the potential of matrix analysis and network theory for the solution
of Power-System problems.
In the early stages, however, the lack of coordination regarding computer
hardware and languages made it difficult to develop techniques for general use
and the tendency was for various supply authorities to develop their own
programs. Later, with the general acceptance of FORTRAN in the Power-
System field, some success was achieved by computer manufacturers in
producing program packages which included load flow, short-circuit and
stability studies.
The dramatic expansion of the subject in the '70's can no longer be recorded in
a single textbook and various specialist treatises are regularly compiled to review
and compare the developments.
While the academic search for new methods continues, the general feeling is
that no major contribution is expected over the basic techniques developed in the
last decade for the modelling of a.c. power systems.
The primary subject of computer modelling is the load-flow problem, which
finds application in all phases of power-system analysis. Due to space limitations,
only the solution of the basic load-flow equations is normally considered. It is
acknowledged, however, that the load-flow problem is not restricted to the
solution of the basic continuously differentiable equations. There is probably not
a single routine program in use anywhere that does not model other features.
Such features often have more influence on convergence than the performance of
the basic algorithm.
The most successful contribution to the load-flow problem has been the
application of Newton-Raphson and derived algorithms. These were finally
established with the development of programming techniques for the efficient
handling of large matrices and in particular the sparsity-oriented ordered-
elimination methods. The Newton algorithm was first enhanced by taking
advantage of the decoupling characteristics of power flow and finally by the use of
reasonable approximations directed towards the use of constant Jacobian
matrices.
In dynamic studies the most significant modelling development has probably
been the application of implicit integration techniques which allow the differen-
tial equations to be algebraized and then incorporated with the network
algebraic equations to be solved simultaneously. The use of implicit trapezoidal
integration has proved to be very stable, permitting step lengths greater than the
smallest time constant of the system. This technique allows detailed repre-
sentation of synchronous machines with their voltage regulators and governors,
induction motors and nonimpedance loads.

1.2. TRANSMISSION SYSTEM DEVELOPMENT

Large increases in transmission distances and voltage levels have resulted in more
sophisticated means of active and reactive power control and the use of d.c.
transmission.
Long-distance transmission presents voltage and power-balancing problems
which can only be accurately assessed by using correspondingly accurate models
of power transmission plant in the phase frame of reference.
Following the first commercial scheme in 1954, h.v.d.c. transmission has been
gradually accepted as a viable alternative to a.c. for many applications and is now
well established.
An all-purpose package of power-system programs can no longer ignore such
developments and the basic algorithms used in conventional a.c. systems need to
be reconsidered for their efficient incorporation.
At first glance the book may appear to overemphasize the role of h.v.d.c.
transmission, while the number of such schemes in existence and under
consideration is still relatively small. However, most of the world's large power
systems contain some large power converters and d.c. transmission links.
Moreover, considering the large power ratings of such schemes, their presence
influences the behaviour of the rest of the system and must be represented
properly.
Whenever possible, any equivalent models used to simulate the converter
behaviour should involve traditional power-system concepts, for easy in-
corporation within existing power-system programs.
However, the number of degrees of freedom of d.c. power transmission is higher
and any attempt to model its behaviour in the more restricted a.c. framework will
have limited application.
The integration of h.v.d.c. transmission with conventional a.c. load-flow
models has been given sufficient coverage in recent years and is now well
understood. Steady-state unbalanced studies, on the other hand, require for more
elaborate modelling in the presence of h.v.d.c. converters and have so far been
given very little consideration.
With reference to power-system disturbances, the behaviour of h.v.d.c.
converters cannot be modelled by conventional fault study or stability programs.
As a result, a.c.-d.c. system planning is normally carried out with the help of
elaborate scaled-down models and simulators. There is considerable reluctance
to the introduction of computer modelling for this purpose on the basis of
inadequate control representation and prohibitive computer runs. However, it is
reasonable to expect that this is only a temporary problem and that in the long
run, computer modelling will become the only practical alternative for the
analysis of a.c.-d.c. system disturbances.

1.3. THEORETICAL MODELS AND COMPUTER PROGRAMS

The dissemination of information on Power System computation takes the form
of technical publications and program instruction manuals.
Technical papers are by necessity concise, their information is mostly
theoretical and their structure schematic. Textbooks, although less restricted,
follow a similar pattern and the extent of practical information normally included
is insufficient to provide adequate understanding of the programs.
Instruction manuals on the other hand, provide full information on the
practical structure of the programs but lack the technical background necessary
for the user to perform the inevitable modifications required in the long run.
It is almost expected that a specialist book of this type should provide a
comprehensive survey and comparison of the various conventional alternatives.
Such approach, although academically satisfying, would detract from the main
object of the book and would occupy invaluable space. Instead, generally
recognized as efficient up-to-date modelling techniques will be described from
theoretical and practical considerations.
Finally, the authors have tried to combine theoretical and practical con-
siderations to introduce in the most direct way the application of such techniques
to the development of a new generation of power-system programs.
 Modelling of Power
Transmission Plant
2.1. INTRODUCTION
Transmission plant components are modelled by their equivalent circuits in
terms of inductance, capacitance and resistance. Each unit constitutes an electric
network in its own right and their interconnection constitutes the transmission
system.
Among the many alternative ways of describing transmission systems to
comply with Kirchhoffs laws, two methods, mesh and nodal analysis, are
normally used. The latter has been found to be particularly suitable for digital
computer work, and is almost exclusively used for routine network calculations.
The nodal approach has the following advantages:
- The numbering of nodes, performed directly from a system diagram, is very
simple.
- Data preparation is easy.
- The number of variables and equations is usually less than with the mesh
method for power networks.
- Network crossover branches present no difficulty.
- Parallel branches do not increase the number of variables or equations.
- Node voltages are available directly from the solution, and branch currents
are easily calculated.
- Off-nominal transformer taps can easily be represented.

This chapter deals with the derivation of equivalent circuits of transmission plant
components and with the formation of the system admittance matrix relating the
current and voltage at every node of the transmission system. This constitutes the
basic framework for the algorithms developed in following chapters.

2.2. LINEAR TRANSFORMATION TECHNIQUES

Linear transformation techniques are used to enable the admittance matrix of
any network to be found in a systematic manner. Consider, for the purposes of
illustration, the network drawn in Fig. 2.1.
Five steps are necessary to form the network admittance matrix by linear
transformation, i.e.
4
 Fig. 2.1. Actual connected network

(i) Label the nodes in the original network.

(ii) Number, in any order, the branches and branch admittances.
(iii) Form the primitive network admittance matrix by inspection.
This matrix relates the nodal injected currents to the node voltages of the
primitive network. The primitive network is also drawn by inspection of the
actual network. It consists of the unconnected branches of the original network
with a current equal to the original branch current injected into the correspond-
ing node of the primitive network. The voltages across the primitive network
branches then equal those across the same branch in the actual network.
The primitive network for Fig. 2.1. is shown in Fig. 2.2.

Fig. 2.2. Primitive or unconnected network

The primitive admittance matrix relationship is:

6
Off-diagonal terms are present where mutual coupling between branches is
present.
(iv) Form the connection matrix [C].
This relates the nodal voltages of the actual network to the nodal voltages of
the primitive network. By inspection of Fig. 2.1,

or in matrix form

(v) The actual network admittance matrix which relates the nodal currents to
the voltages by,

can now be derived from,

C Yak1 = Ccl 'C Y,,l Ccl
3x3 3x5 5k5 5x3
which is a straightforward matrix multiplication.

2.3. BASIC SINGLE-PHASE MODELLING

Under perfectly balanced conditions, transmission plant can be represented by
single-phase models, the most extensively used being the equivalent-lr circuit.
Transmission lines
In the case of a transmission line the total resistance and inductive reactance of
the'line is included in the series arm of the n-equivalent and the total capacitance
to neutral is divided between its shunt arms.

Transformer on nominal ratio

The equivalent-n model of a transformer is illustrated in Fig. 2.3, where yo, is the
reciprocal of zoc(magnetizing impedance) and yscis the reciprocal of zsc(leakage
impedance). zscand zocare obtained from the standard short-circuit and open-

p r 1[
circuit tests.
This yields the following matrix equation:

- Y s c + Yoclz
where
ys, is the short-circuit or leakage admittance,
yo, is the open-circuit or magnetizing admittance.
The use of a three-terminal network is restricted to the single-phase
representation and cannot be used as a building block for modelling three-phase
transformer banks.
The magnetizing admittances are usually removed from the transformer model
and added later as small shunt-connected admittances at the transformer
terminals. In the per unit system the model of the single-phase transformer can
then be reduced to a lumped leakage admittance between the primary and
secondary busbars.

Off-nominal transformer tap settings

A transformer with turns ratio 'a' interconnecting two nodes i,k can be
represented by an ideal transformer in series with the nominal transformer
leakage admittance as shown in Fig. 24a).

Fig. 2.3. Transformer equivalent circuit

If the transformer is on nominal tap (a = I), the nodal equations for the network
branch in the per unit system are:
Iik = Y i k vi - Y i k vk (2.3.2)
Iki =yik vk - Y i k vi (2.3.3)
In this case Iik = - Iki.
For an off-nominal tap setting and letting the voltage on the k side of the ideal
transformer be Vt we can write

Eliminating Vt between equations (2.3.4) and (2.3.5) we obtain

A simple equivalent 7c circuit can be deduced from equations (2.3.7) and (2.3.8) the
elements of which can be incorporated into the admittance matrix. This circuit is
illustrated in Fig. 2.4(b)

(a (b)
Fig. 2.4. Transformer with off-nominal tap setting

The equivalent circuit of Fig. 2.4(b) has to be used with care in banks
containing delta-connected windings. In a star-delta bank of single-phase
transformer units, for example, with nominal turns ratio, a value of 1.0 per unit
voltage on each leg of the star winding produces under balanced conditions 1.732
per unit voltage on each leg of the delta winding (rated line to neutral voltage as
base). The structure of the bank requires in the per unit representation an effective
tapping at fi nominal turns ratio on the delta side, i.e. a = 1.732.
For a delta-delta or star-delta transformer with taps on the star winding, the
equivalent circuit of Fig. 2.4(b) would have to be modified to allow for effective
taps to be represented on each side. The equivalent-circuit model of the single-
phase unit can be derived by considering a delta-delta transformer as comprising
a delta-star transformer connected in series (back to back) via a zero-impedance
link to a star-delta transformer, i.e. star windings in series. Both neutrals are
solidly earthed. The leakage impedance of each transformer would be half the
impedance of the equivalent delta-delta transformer. An equivalent per unit
representation of this coupling is shown in Fig. 2.5. Solving this circuit for
terminal currents :
I' (V' - V " ) ,
Ip=-=
a a

Fig. 2.5. Basic equivalent circuit in p.u. for coupling between primary and
secondary coils with both primary and secondary off-nominal tap ratios of a and B

or in matrix form:

These admittance parameters form the primitive network for the coupling
between a primary and secondary coil.

Phase-shifting transformers
To cope with phase-shifting, the transformer of Fig. 2.5 has to be provided with a
complex turns ratio. Moreover, the invariance of the product VI* across the ideal
transformer requires a distinction to be made between the turns ratios for current
and voltage, i.e.
Thus the circuit of Fig. 2.5 has two different turns ratios, i.e.
a, = a +jb for the voltages
and
a,= a -jb for the currents
Solving the modified circuit for terminal currents:

Thus, the general single-phase admittance of a transformer including phase

shifting is:
I I I

It should be noted that although an equivalent lattice network similar to that

in Fig. 2.5 could be constructed, it is no longer a bilinear network as can be seen
from the asymmetry of y in equation (2.3.14). The equivalent circuit of a
single-phase phase-shifting transformer is thus of limited value and the trans-
former is best represented analytically by its admittance matrix.

2.4. THREE-PHASE SYSTEM ANALYSIS

Discussion of the frame of reference

Sequence components have long been used to enable convenient examination of
the balanced power system under both balanced and unbalanced loading
conditions.
The symmetrical component transformation is a general mathematical
technique developed by Fortescue whereby any 'system of n vectors or quantities
may be resolved, when n is prime, into n different symmetrical n phase systemsy.(')
Any set of three-phase voltages or currents may therefore be transformed into
three symmetrical systems of three vectors each. This, in itself, would not
commend the method and the assumptions, which lead to the simplifying nature
of symmetrical components, must be examined carefully.
Consider, as an example, the series admittance of a three-phase transmission
line, shown in Fig. 2.6, i.e. three mutually coupled coils. The admittance matrix
relates the illustrated currents and voltages by
C1I"k = CY1,kCvlak (2.4.1)
where

and

By the use of the symmetrical component transformation the three cobs of

Fig. 2.6 can be replaced by three uncoupled coils. This enables each coil to be
treated separately with a great simplification of the mathematics involved in the
analysis.

Fig. 2.6. Admittance representation of a

three-phase series element. (a) Series admit-
tance element; (b)Admittance matrix
representation
 The transformed quantities (indicated by subscripts 012 for the zero, positive
and negative sequences respectively) are related to the phase quantities by:

where [T,] is the transformation matrix.

The transformed voltages and currents are thus related by the transformed
admittance matrix,

Assuming that the element is balanced, we have

and a set of invariant matrices [TI exist. Transformation (2.4.6) will then yield a
diagonal matrix [y],, ,.
In this case the mutually coupled three-phase system has been replaced by
three uncoupled symmetrical systems. In addition, if the generation and loading
are balanced, or may be assumed balanced, then only one system, the positive
sequence system, has any current flow and the other two sequences may be
ignored. This is essentially the situation with the single-phase load flow.
If the original phase admittance matrix [yak] is in its natural unbalanced state
then the transformed admittance matrix [yo, ,] is full. Therefore, current flow of
one sequence will give rise to voltages of all sequences, i.e. the equivalent circuits
for the sequence networks are mutually coupled. In this case the problem of
analysis is no simpler in sequence components than in the original phase
components and symmetrical components should not be used.
From the above considerations it is clear that the asymmetry inherent in all
power systems cannot be studied with any simplification by using the symmetri-
cal component frame of reference. Data in the symmetrical component frame
should only be used when the network element is balanced, for example,
synchronous generators.
In general, however, such an assumption is not valid. Unsymmetrical
interphase coupling exists in transmission lines and to a lesser extent in
transformers and this results in coupling between the sequence networks.
Furthermore, the phase shift introduced by transformer connections is difficult to
represent in sequence component models.
With the use of phase coordinates the following advantages become apparent:
- Any system element maintains its identity.
- Features such as asymmetric impedances, mutual couplings between phases
 and between different system elements, and line transpositions are all readily
considered.
- Transformer phase shifts present no problem.

The use of compound admittances

When analysing three-phase networks, where the three nodes at a busbar are
always associated together in their interconnections, the graphical representation
of the network is greatly simplified by means of 'compound admittances', a
concept which is based on the use of matrix quantities to represent the
admittances of the network.
The laws and equations of ordinary networks are all valid for compound
networks by simply replacing single quantities by appropriate matrices.")
Consider six mutually coupled single admittances, the primitive network of
which is illustrated below in Fig. 2.7.

Fig. 2.7. Primitive network of six coupled admittances

The primitive admittance matrix relates the nodal injected currents to the
branch voltages as follows:

Partitioning equation (2.4.8) into 3 x 3 matrices and 3 x 1 vectors, the

equation becomes,
where

Graphically we represent this partitioning as grouping the six coils into two
compound coils (a) and (b), each composed of three individual admittances. This
is illustrated in Fig. 2.8.

Fig. 2.8. Two coupled compound admittances

On examination of [Yab] and [Yba] it can be seen that,

if, and only if yik = yki for i = 1 to 3 and k = 4 to 6. That is, if and only if the
couplings between the two groups of admittances are bilateral.
 In this case equation (2.4.9) may be written

The primitive network for any number of compound admittances is formed in

exactly the same manner as for single admittances, except in that all quantities are
matrices of the same order as the compound admittances.
The actual admittance matrix of any network composed of the compound
admittances can be formed by the usual method of linear transformation; the
elements of the connection matrix are now n x n identity matrices where n is the
dimension of the compound admittances.
If the connection matrix of any network can be partitioned into identity
elements of equal dimensions greater than one, the use of compound admittances
is advantageous.
As an example, consider the network shown in Figs. 2.9 and 2.10, this
represents a simple line section. The admittance matrix will be derived using
single and compound admittances to show the simple correspondence. The
primitive networks and associated admittance matrices are drawn in Fig. 2.11.
The connection matrices for the single and compound networks are illustrated by
equations (2.4.12) and (2.4.13) respectively.
Fig. 2.9. Sample network represented by single admittances

Fig. 2.10. Sample network represented by com-

pound admittances

(ii)
 (iv)
Fig. 2.1 1. Primitive networks and corresponding admittance matrices.
(i) Primitive network using single admittances; (ii) Primitive admittance matrix ;
(iii) Primitive network using compound admittances; (iv) Primitive admittance
matrix

The exact equivalence, with appropriate matrix partitioning, is clear.

The network admittance matrix is given by the linear transformation equation,
CYNODEI = C C I T C Y ~ d C C l
This matrix multiplication can be executed using the full matrices or in
partitioned form. The result in partitioned form is,

Rules for forming the admittance matrix of simple networks

The method of linear transformation may be used to obtain the admittance
matrix of any network. For the special case of networks where there is no mutual
coupling, simple rules may be used to form the admittance matrix by inspection.
These rules, which apply to compound networks with no mutual coupling
between the compound admittances, may be stated as follows:
(a) Any diagonal term is the sum of the individual branch admittances
connected to the node corresponding to that term.
 (b) Any off-diagonal term is the negated sum of the branch admittances which
are connected between the two corresponding nodes.

Network subdivision
To enable the transmission system to be modelled in a systematic, logical and
convenient manner the system must be subdivided into more manageable units.
These units, called subsystems, are defined as follows: A SUBSYSTEM is the unit
into which any part of the system may be divided such that no subsystem has any
mutual couplings between its constituent branches and those of the rest of the
system. This definition ensures that the subsystems may be combined in an
extremely straightforward manner.
The system is first subdivided into the most convenient subsystems consistent
with the definition above.
The most convenient unit for a subsystem is a single network element. In
previous sections the nodal admittance matrix representation of all common
elements has been derived.
The subsystem unit is retained for input data organization. The data for any
subsystem is input as a complete unit, the subsystem admittance matrix is
formulated and stored and then all subsystems are combined to form the total
system admittance matrix.

2.5. THREE-PHASE MODELS OF TRANSMISSION LINES

Transmission line parameters are calculated from the line geometrical character-
istics. The calculated parameters are expressed as a series impedance and shunt
admittance per unit length of line. The effect of ground currents and earth wires
are included in the calculation of these parameter^'^"^).
SERIES IMPEDANCE
A three-phase transmission line with a ground wire is illustrated in Fig. 2.12(a).
The following equations can be written for phase 'a':

and substituting
1, = I,+ 1, + 1, + I,
Va - Vi = Ia(Ra +joL,) + I,joL,, + I,joL,,
+jwL,;I, -joL,,(I, + I, + I, + I,) + V,
Regrouping and substituting for V,, i.e.
 19
+ I,(jwL,, - jwL,, + R, +jwL, -jwL,,)
+ IC(jojL, -jwL,, + R, +jwL, -jwL,)
+ I,O'wL, -jwL,, + R, +jwL, -jwL,,)
+
AVa = Ia(Ra jwL, - 2jwLa, + R, +joL,)
+ Ib(jwLab-jwL,, -jwL,, + R, +jwL,)
+ Ic(jwL, -jwLc, - jwL,, + R, +jwL,)
+ Ig(jwLag-jwL,, -jwL,, + R, +jwL,)
or
A v a = zaa- +
,Ia + Zab-,Ib ZaC-,Ic Zag-,I#+ (2.5.1)
and writting similar equations for the other phases the following matrix equation
results:

Fig. 2.12. (a) Three-phase transmission series im-

pedance equivalent; (b) Three-phase transmission
shunt impedance equivalent
 Since we are interested only in the performance of the phase conductors it is
more convenient to use a three-conductor equivalent for the transmission line.
This is achieved by writing Matrix equation (2.5.2)in partitioned form as follows:

From (2.5.3).

From equations (2.5.3) and (2.5.4) and assuming that the ground wire is at zero
potential :

SHUNT ADMI'ITANCE

With reference to Fig. 2.12(b) the potentials of the line conductors are related to
the conductor charges by the matrix e q ~ a t i o n ' ~ ) :

Similar considerations as for the series impedance matrix lead to

where PA, is a 3 x 3 matrix which includes the effects of the ground wire. The
capacitance matrix of the transmission line of Fig. 2.12 is given by
 The series impedance and shunt admittance lumped-.rr model representation of
the three-phase line is shown in Fig. 2.13(a)and its matrix equivalent is illustrated
in Fig. 2.13(b).These two matrices can be represented by compound admittances,
(Fig. 2.13(c)) as described earlier.

t-blf shunt
admittance

Fig. 2.13. Lumped-n model of a short three-phase line series impedance. (a) Full
circuit representation; (b) Matrix equivalent; (c) Using three-phase compound
admittances
 Following the rules developed for the formation ofthe admittance matrix using
the compound concept, the nodal injected currents of Fig. 2.13(c) can be related to
the nodal voltages by the equation,

This forms the element admittance matrix representation for the short line
between busbars i and k in terms of 3 x 3 matrix quantities.
This representation may not be accurate enough for electrically long lines. The
physical length at which a line is no longer electrically short depends on the
wavelength, therefore if harmonic frequencies are being considered, this physical
length may be quite small. Using transmission line and wave propagation theory
more exact models may be However, for normal mains frequency
qnalysis, it is considered sufficient to model a long line as a series of two or three
nominal-n sections.

Mutually coupled three-phase lines

When two or more transmission lines occupy the same right of way for a
considerable length, the electrostatic and electromagnetic coupling between
those lines must be taken into account.
Consider the simplest case of two mutually coupled three-phase lines. The two
coupled lines are considered to form one subsystem composed of four system
busbars. The coupled lines are illustrated in Fig. 2.14, where each element is a
3 x 3 compound admittance and all voltages and currents are 3 x 1 vectors.
The coupled series elements represent the electromagnetic coupling while the
coupled shunt elements represent the capacitive or electrostatic coupling. These
coupling parameters are lumped in a similar way to the standard line parameters.
With the admittances labelled as in Fig. 2.14 and applying the rules of linear
transformation for compound networks the admittance matrix for the subsystem
is defined as follows:

12 x 1 12 x 12 12x 1
It is assumed here that the mutual coupling is bilateral. Therefore.Y2, = Y T2
etc.
The subsystem may be redrawn as Fig. 2.15. The pairs of coupled 3 x 3
compound admittances are now represented as a 6 x 6 compound admittance.
The matrix representation is also shown. Following this representation and the
labelling of the admittance blocks in the figure, the admittance matrix may be
written in terms of the 6 x 6 compound coils as,

(ii)
Fig. 2.15. 6 x 6 compound admittance representation of two coupled three-
phase lines. (i) 6 x 6 Matrix representation; (ii) 6 x 6 Compound admittance
representation
 This is clearly identical to equation (2.5.10) with the appropriate matrix
partitioning.
The representation of Fig. 2.15 is more concise and the formation of equation
(2.5.11 ) from this representation is straightforward, being exactly similar to that
which results from the use of 3 x 3 compound admittances for the normal single
three-phase line.
The data which must be available, to enable coupled lines to be treated in a
similar manner to single lines, is the series impedance and shunt admittance
matrices. These matrices are of order 3 x 3 for a single line, 6 x 6 for two coupled
lines, 9 x 9 for three and 12 x 12 for four coupled lines.
Once the matrices [Z,] and [Y,] are available, the admittance matrix for the
subsystem is formed by application of equation (2.5.11).
When all the busbars of the coupled lines are distinct, the sub-system may be
combined directly into the system admittance matrix. However, if the busbars are
not distinct then the admittance matrix as derived from equation (2.5.11)must be
modified. This is considered in the following section.

Consideration of terminal connections

The admittance matrix as derived above must be reduced if there are different
elements in the subsystem connected to the same busbar. As an example consider
two parallel transmission lines as illustrated in Fig. 2.16.

Fig. 2.16. Mutually coupled parallel transmission lines

The admittance matrix derived previously related the currents and voltages at
the four busbar A l , A2, B1 and B2. This relationship is given by:
 25
The nodal injected current at busbar A,(',), is given by,
A
' = 'A1 + 'A2
similarly
B
' =IBl +' ~ 2
Also from inspection of Fig. 2.16,
VA= VAl = VA2
VB= VB1 = VB1

The required matrix equation relates the nodal injected currents, 1, and I,, to
the voltages at these busbar. This is readily derived from equation (2.5.12)and the
conditions specified above. This is simply a matter of adding appropriate rows
and columns and yields,

This matrix [YAB] is the required nodal admittance matrix for the subsystem.
It should be noted that the matrix in equation (2.5.12) must be retained as it is
required in the calculation of the individual line power flows.

Shunt elements t

Shunt reactors and capacitors are used in a power system for reactive power
control. The data for these elements are usually given in terms of their rated MVA
and rated kV; the equivalent phase admittance in p.u. is calculated from these
data.
Consider, as an example, a three-phase capacitor bank shown in Fig. 2.17. A
similar triple representation as that for a line section is illustrated. The final two
forms are the most compact and will be used exclusively from this point on.

Fig. 2.17. Representation of a shunt capacitor bank

 The admittance matrix for shunt elements is usually diagonal as there is
normally no coupling between the components of each phase. This matrix is then
incorporated directly into the system admittance matrix, contributing only to the
self admittance of the particular bus.

Series elements
Any element connected directly between two buses may be considered a series
element. Series elements are often taken as being a section in a line sectional-
ization which is described later in the chapter.
A typical example is the series capacitor bank which is usually taken as
uncoupled, i.e. the admittance matrix is diagonal.
This can be represented graphically as in Fig. 2.18.
The admittance matrix for the subsystem can be written by inspection as:

1- S E I I

Fig. 2.1 8. Graphic representation of series capacitor bank between nodes i and k

2.6. THREEPHASE MODELS OF TRANSFORMERS

The inherent assumption, that the transformer is a balanced three-phase device, is
justified in the majority of practical situations, and traditionally, t hree-phase
transformers are represented by their equivalent sequence networks.
More recently, however, methods have been d e ~ e l o p e d ( ~to
) . (enable
~ ) all three-
phase transformer connections to be accurately modelled in phase coordinates.
In phase coordinates no assumptions are necessary although physically justifi-
able assumptions are still used in order to simplify the model. The primitive
admittance matrix, used as a basis for the phase coordinate transformer model is
derived from the primitive or unconnected network for the transformer windings
and the method of linear transformation enables the admittance matrix of the
actual connected network to be found.
Primitive admittance model of three-phase transformers
Many three-phase transformers are wound on a common core and all windings
are therefore coupled to all other windings. Therefore, in general, a basic two-
winding three-phase transformer has a primitive or unconnected network
consisting of six coupled coils. If a tertiary winding is also present the primitive
network consists of nine coupled coils. The basic two-winding transformer shown
in Fig. 2.19 is now considered, the addition of further windings being a simple but
cumbersome extension of the method.
The primitive network, Fig. 2.20, can be represented by the primitive
admittance matrix which has the following general form:

Fig. 2.19. Diagrammatic representation of twewinding transformer

Fig. 2.20. Primitive network of two-winding transformer. Six

coupled coil primitive network. (Note the dotted coupling
represents parasitic coupling between phases)
28
The elements of matrix [Y] can be measured directly, i.e. by energizing coil i
and short-circuiting ail other coils, column i of [Y] can be calculated from
yki = IJVi.
Considering the reciprocal nature of the mutual couplings in equation (2.6.1)
twenty-one short circuit measurements would be necessary to complete the
admittance matrix. Such a detailed representation is seldom required.
By assuming that the flux paths are symmetrically distributed between all
windings equation (2.6.1) may be simplified to equation (2.6.2).

where
yk is the mutual admittance between primary coils;
yk is the mutual admittance between primary and secondary coils on
different cores ;
yz is the mutual admittance between secondary coils.
For three separate single-phase units all the primed values are effectively zero.
In three-phase units the primed values, representing parasitic interphase
coupling, do have a noticeable effect. This effect can be interpreted through the
symmetrical component equivalent circuits.

where y , = y l a f , y,, = and M i j= ylaiPi

for i = 1,2 or 3 a n d j = 4 , 5 or 6
Fig. 2.21. Primitive network
 If the values in equation (2.6.2) are available then this representation of the
primitive network should be used. If interphase coupling can be ignored, the
coupling between a primary and a secondary coil is modelled as for the single-
phase unit, giving rise to the primitive network of Fig. 2.21.
The new admittance matrix equation is

Models for common transformer connections

The network admittance matrix for any two-winding three-phase transformer
can now be formed by the method of linear transformation.
As a simple example, consider the formation of the admittance matrix for a
star-star connection with both neutrals solidly earthed in the absence of
interphase mutuals. This example is chosen as it is the simplest computationally.
The connection matrix is derived from consideration of the actual connected
network. For the star-star transformer illustrated in Fig. 2.22, the connection
matrix [C] relating the branch voltages (i.e. voltages of the primitive network) to
the node voltages (i.e. voltages of the actual network) is a 6 x 6 identity matrix, i.e.
 Fig. 2.22. Network connection diagram for three-phase star-star transformer

The nodal admittance matrix [YJNODEis given by,

CYI,OD, = CCItCY1PR1,CCI
Substituting for [C] yields,

Let us now consider the Wye G-Delta connection illustrated in Fig. 2.23.

Fig. 2.23. Network connection diagram for Wyi? G-Delta transformer

The following connection can be written by inspection between the primitive

branch voltages and the node voltages
 and using [YIPRIMfrom equation (2.6.2).

Moreover, if the primitive admittances are expressed in per unit, with both the
primary and secondary voltages being one per unit, the Wye-Delta transformer
. model must include an effective turns ratio offi. The upper right and lower left
quadrants of matrix (2.6.10)must be divided by fi and the lower right quadrant
by 3.
In the particular case of three-single phase transformer units connected in Wye
G-Delta all the y' and y" terms will disappear. Ignoring off nominal taps (but
keeping in mind the effective fi turns ratio in per unit) the nodal admittance
matrix equation relating the nodal currents to the nodal voltages is

where Y is the transformer leakage admittance in p.u.

An equivalent circuit can be drawn, corresponding to this admittance model of
the transformer, as illustrated in Fig. 2.24.
The large shunt admittances to earth from the nodes of the star connection are
apparent in the equivalent circuit. These shunts are typically around 10 p.u. (for a
10 percent leakage reactance transformer).
 Fig. 2.24. Equivalent circuit for Star-Delta transformer

The models for the other common connections can be derived following a
similar procedure.
In general, any two-winding three-phase transformer may be represented using
two coupled compound coils. The network and admittance matrix for this
representation is illustrated in Fig. 2.25.

Fig. 2.25. Two winding threephase transformer as two

coupled compound coils

It should be noted that,

as the coupling between the two compound coils is bilateral.

Often, because more detailed information is not required, the parameters of all
three phases are assumed balanced. In this case the common three-phase
connections are found to be modelled by three basic submatrices.
The submatrices, [Y,,], [Y,] etc., are given in Table 2.1 for the common
connections.

Table 2.1. Characteristic submatrices used in forming the transformer admittance matrices

Trans connection Self admittance Mutual admittance

Bus P Bus S YPP yss

Wye-G Wye - G YI YI
Wye - G WY~ y1~,3 y ~3 ~ /
Wye-G Delta YI YII
W ~ e W ~ e y11/3 y11,3
WY~ Wye y1113 y11
Delta Delta YII Yrl

Basic subrnatrices used in node admittance formulation of common three-phase transformer

connections, where:

Finally, these submatrices must be modified to accounts for off-nominal tap

ratio as follows:
(i) Divide the self admittance of the primary by a2.
(ii) Divide the self admittance of the secondary by P2.
(iii) Divide the mutual admittance matrices by (ap).
It should be noted that in the p.u. system a delta winding has an off-nominal tap
of Js.
 For transformers with ungrounded Wye connections, or with neutrals
connected through an impedance, an extra coil is added to the primitive network
for each unearthed neutral and the primitive admittance matrix increases in
dimension. By noting that the injected current in the neutral is zero, these extra
terms can be eliminated from the connected network admittance matrix.(7)
Once the admittance matrix has been formed for a particular connection it
represents a simple subsystem composed of the two busbars interconnected by
the transformer.

Three-phase transformer models with independent phase tap control

Disregarding interphase mutual couplings the per unit primitive admittance
matrix in terms of the transformer leakage admittance (y,,) is:

where a,, a, and a, are the off-nominal taps on windings 1,2, and 3 respectively.
In addition any windings connected in delta will, because of the per unit system,
have an effective tap offi.
The nodal admittance matrix for the transformer windings is:

where LC] is the connection (windings to nodes) matrix.

As an example [Yn,,,] for a star-delta transformer with earthed neutral is as
follows :
 - Ytl
a a 1

Sequence components modelling of three-phase transformers

In most cases lack of data will prevent the use of the general model based on the
primitive admittance matrix and will justify the conventional approach in terms
of symmetrical components. Let us now derive the general sequence components
equivalent circuits and the assumptions introduced in order to arrive at the
conventional models.
With reference to the Wye G-Delta common-core transformer of Fig. 2.23,
represented by equation (2.6.10), and partioning this matrix to separate self and
mutual elements the following transformations apply:
Primary side :

where
36
Therefore

Secondary side
The delta connection on the secondary side introduces an effective ,,hturns ratio
and the sequence components admittance matrix is

Mutual terms
The mutual admittance submatrix of equation (2.6.10), modified for effective
turns ratio is transformed as follows:
 37
Recombining the sequence components submatrices yields

Equation (2.6.15)can be represented by the three sequence networks of Figs. 2.26,

2.27 and 2.28 respectively.
Wve G
Yc + 2YID) Delta

Fig. 2.26. Zero-sequence node admittance model for

a commoncore grounded Wye-Delta transformer
(3). ( 0 1982 IEEE)

Fig. 2.27. Positive-sequence node admittance model for a commoncore grounded

Wye-Delta Transformer (3). ( 0 1982 IEEE)

Wye G (Y, +Y," )/-30° Delta

Fig. 2.28. Negative-sequence node admittance model for a commoncore

grounded Wye-Delta transformer (3). ( 0 1982 IEEE)
 In general, therefore, the three sequence impedances are different on a
common-core transformer.
The complexity of these equivalent models is normally eliminated by the
following simplifications:
- The 30" phase shifts of Wye-Delta connections are ignored.
- The interphase mutuals admittances are assumed equal, i.e. yk = yk = y.:
These are all zero with uncoupled single-phase units.
- The differences (y, - y,) and (y, - y,) are very small and are therefore
ignored.
With these simplifications, Table 2.2. illustrates the sequence impedance
models of three-phase transformers in conventional steady-state balanced
transmission system studies.

Table 2.2. Typical symmetricalcomponent models for the six most

common connections of three-phase transformers (3). (01982
IEEE)

Bus P Bus 0 Pos Seq Zero Srq

p 'sc
Wye G Wye G w0

'sc
Wye G w P
J'-

Wye G Delta

W Y ~ W Y ~

w e blta

Delta Delta

Line sectionalization
A line may be divided into sections to account for features such as the following:
- Transposition of line conductors.
- Change of type of supporting towers.
- Variation of soil permitivity.
 39
- Improvement of line representation. (Series of two or more equivalent n
networks.)
- Series capacitors for line compensation.
- Lumping of series elements not central to a particular study.
An example of a line divided into a number of sections is shown in Fig. 2.29.

Section 4 Section6
Bus A PI P2
Section I Section Section
1 : 2 j 3

r
I I
I
1
0

11 II 4
I\
b
1 II ti

I
Transposition /
Change of
Series &aci+ors
obc obc
Phases configuration

Fig. 2.29. Example of a transmission line divided into sections

The network of Fig. 2.29 is considered to form a single subsystem. The

resultant admittance matrix between bus A and bus B may be derived by finding,
for each section, the ABCD or transmission parameters, then combining these by
matrix multiplications to give the resultant transmission parameters. These are
then converted to the required admittance parameters.
This procedure involves an extension of the usual two port network theory to
multi-two-port networks. Currents and voltages are new matrix quantities and
are defined in Fig. 2.30. The ABCD matrix parameters are also shown.
The dimensions of the parameters matrices correspond to those of the section
being considered, i.e. 3, 6, 9, or 12 for 1,2, 3 or 4 mutually coupled three-phase
elements respectively. All sections must contain the same number of mutually
coupled three-phase elements, ensuring that all the parameter matrices are of the
same order and that the matrix multiplications are executable. To illustrate this
feature, consider the example of Fig. 2.31.

FEATURES OF INTEREST

As a matter of programming convenience an ideal transformer is created

and included in Section I.
The dotted coupling represents coupling which is zero. It is included to
ensure correct dimensionality of all matrices.
In the p.u. system the mutual coupling between the 220-kV and 66-kV lines
is expressed to a voltage base given by the geometric mean of the base line-
neutral voltages of the two parallel circuits.
 Fig.2.30. Two port network transmission parameters.
(i) Normal two port network; (ii) Transmission parameters;
(iii) Multi-two port network; (iv) Matrix transmission
parameters

2 2 0 kV
22O/66 kV

Bus A 4 220/66 kV 3 (i) 6 6

Bus CkV

Bus B

Bus A
2 2 0 kV

I I
I
Section No1 ! ~ e c t i o n ~ o 2Section
! No 3 1
( ii 1
Fig. 2.31. Sample system to illustrate line sectionalization.
(i) System single line diagram;(ii) system redrawn to illustrate line
sectionalization
 Table 2 . 3 . ABCD Parameter matrices for the common section types

Cul + CZI C YIP - CZI

Transmission line
C Yl { Cul + CZI C ~ 1 1 4 ) - { Cul + CYI CZIl21

- CYSPI -'CYSSI CYSPI - '

Transformer
CYPSI -CyPP1C~sP1-'C~ss1 CYPPICYSPI-'

Shunt element
I [.I [OI I
Series element
i ;; I-;;J
In Table 2.3 [u] is the unit matrix, [O] is a matrix of zeros, and all other matrices
have been defined in their respective sections.
Note, all the above matrices have dimensions corresponding to the number of
coupled three-phase elements in the section.
Once the resultant ABCD parameters have been found the equivalent nodal
admittance matrix for the subsystem can be calculated from the following
equation.

CYl =
I [Dl CBI - '

CBI - '
CCI - [Dl CBI - ' CAI

'
- CBI - CAI

2.7. FORMATION O F THE SYSTEM ADMITTANCE MATRIX

It has been shown that the element (and subsystem) admittance matrices can be
manipulated efficiently if the three nodes at the busbar are associated together.
This association proves equally helpful when forming the admittance matrix for
the total system.
The subsystem, as defined in Section 2.4., may have common busbars with
other subsystems, but may not have mutual coupling terms to the branches of
other subsystems. Therefore the subsystem admittance matrices can be combined
to form the overall system admittance matrix as follows:
- The self-admittance of any busbar is the sum of all the individual self-
admittance matrices at that busbar.
- The mutual-admittance between any two busbars is the sum of the individual
mutual-admittance matrices from all the subsystems containing those two
nodes.

2.8. REFERENCES
1. E. Clarke, 1943. Circuit Analysis of A.C. Power Systems, Vol. I. Wiley and Sons Ltd.,
New York.
2. G. Kron, republished in 1965. Tensor Analysis of Networks. MacDonald, London.
3. M. S. Chen and W. E. Dillon, July 1974. 'Power-system modelling', Proc. IEEE, 62 (7),
901.
4. M. A. Laughton, 1968. 'Analysis of unbalanced polyphase networks by the method of
phase coordinates, Part I. System representation in phase frame of reference', Proc.
IEE, 115 (8), 1 163- 1 172.
5. K. J. Bowman and J. M. McNamee, 1964. 'Development of equivalent pi and T matrix
circuits for long untransposed transmission lines', IEEE, PAS-84, 625-632.
6. L. M. Wedepohl and R. G. Wasley, 1966. 'Wave propagation in multiconductor
overhead lines', Proc. IEE, 113 (4), 627-632.
7. W. E. Dillon and M. S. Chen, 1972. 'Transformer modelling in unbalanced three-phase .
networks.' IEEE Summer Power Meeting, Vancouver.
 Modelling of Static A C - D . C .
Conversion Plant
3.1. INTRODUCTION
Although the power electronic device is basically a switch, it is only explicitly
represented as such in dynamic studies (Chapters 11 and 12). The periodicity of
switching sequences can be used in steady-state studies to model the active and
reactive power loading conditions of a.c.-d.c. converters at the relevant busbars.
Such modelling is discussed here with reference to the most common con-
figuration used in power systems, i.e. the three-phase bridge rectifier shown in
Fig. 3.1.

Fig. 3.1. Basic three-phase rectifier bridge

For large power ratings static converter units generally consist of a number of
series and/or parallel connected bridges, some or all bridges being phase-shifted
relative to the others. With these configurations twelve-pulse and higher pulse
numbers can be achieved to reduce the distortion of the supply current with
limited or no filtering. A multiple bridge rectifier can therefore be modelled as a
single equivalent bridge with a sinusoidal supply voltage at the terminals.
The following basic assumptions are normally made in the development of the
steady-state m ~ d e l ( ' ) . ( ~ ) . ( ~ ) . ( ~ ) :
(i) The forward voltage drop in a conducting valve is neglected so that the
valve may be considered as a switch. This is justified by the fact that the
voltage drop is very small in comparison with the normal operating
 \
voltage. It is, further, quite independent of the current and should,
therefore, play an insignificant part in the commutation process since all
valves commutating on the same side of the bridge suffer similar drops.
Such a voltage drop is taken into account by adding it to the d.c. line
resistance. The transformer windings resistance is also ignored in the
development of the equations, though it should also be included to
calculate the power loss.
(ii) The converter transformer leakage reactances as viewed from the secon-
dary terminals are identical for the three phases, and variations of leakage
reactance caused by on-load tap-changing are ignored.
(iii) The direct current ripple is ignored, i.e. sufficient smoothing inductance is
assumed on the d.c. side.

3.2. RECTIFICATION
Rectifier loads can use diode and thyristor elements in full or half-bridge
configurations. In some cases the diode bridges are complemented by transfor-
mer on-load tap-changer and saturable reactor control. Saturable reactors
produce the same effect as thyristor control over a limited range of delay angles

Fig. 3.2. Diode rectifier waveforms. (a)

Alternating current in phase 'b'; (b) Common
anode (ca) and cathode (cc) voltage waveform;
(c) Rectified voltage
 Referring to the voltage waveforms in Fig. 3.2. and using as time reference the
instant when the phase to neutral voltage in phase 'b' is a maximum, the
commutating voltage of valve 3 can be expressed as:

e, - e, = JIav,,,, sin (at + t)

where 'a' is the off-nominal tap-change position of the converter transformer. The
shaded area in Fig. 3.2(b) indicates the potential difference between the common
cathode (cc) and common anode (ca) bridge poles for the case of uncontrolled
rectification. The maximum average rectified voltage is therefore:

v = a v sin (3.2.1)
43
0

However, uncontrolled rectification is rarely used in large power conversion.

Controlled rectification is achieved by phase-shifting the valve conducting
periods with respect to their corresponding phase voltage waveforms.

Fig. 3.3. Thyristorcontrolled waveform. (a) Alternating current in

phase 'b'; (b) Rectified d.c. voltage wafeforms

With delay angle control the average rectified voltage (shown in Fig. 3.3) is thus

In practice the voltage waveform is that of Fig. 3.4, where a voltage area (6A)is
lost due to the reactance (X,) of the a.c. system (as seen from the converter),
 Fig. 3.4. Effect of commutation reactance. (a) Alternating cur-
rent; (b) D.c. voltage wavefoms

referred to as commutation reactance. The energy stored in this reactance has to

be transferred from the outgoing to the incoming phase, and this process results
in a commutation or conduction overlap angle (u). Referring to Fig. 3.4, and
ignoring the effect of resistance in the commutation circuit, area 6 A can be
determined as follows:

where e,, e, are the instantaneous voltages of phases a and b respectively, and i, is
the incoming valve (commutating) current.

Finally, by combining equations (3.2.1), (3.2.2) and (3.2.4)the following a.c.-d.c.

voltage relationship is obtained
6A 3J2 3
vd= vOcos~--=-
n/3 n
aVtermcos a - -XcId
n
(3.2.5)

It must be emphasized that the commutating voltage (Vterm) is the a.c. voltage
at the closest point to the converter bridge where sinusoidal waveforms can be
assumed. The commutation reactance (X,) is the reactance between the point at
which Vtermexists and the bridge. Where filters are installed the filter busbar
voltage can be used as Vterm.In the absence of filters, Vtermmust be established at
some remote point and X, must be modified to include the system impedance
from the remote point to the converter.
 With perfect filtering, only the fundamental component of the current
waveform will appear in the a.c. system. This component is obtained from the
Fourier analysis of the current waveform in Fig. 3.4, and requires information
of i, and u.
Taking as a reference the instant when the line voltage (e, - e,) is zero, equation
(3.2.3) can be written as
X, di
$a. Vtermsin o t = 2 -2
o dt
and integrating with respect to (ot):

1
- -a. Vtermcos ot + K = X,i,
fi
From the initial condition: i, = 0 at ot = a the following expressions for K and
i, are obtained.

From the final condition: i, = I , at ot = a + u the following expressions for I , and

u are obtained.
a . Vterm [cos a - cos (a + u)]
Id = -
Jzxc

Equation (3.2.6) provides the time-varying commutating current and equation

(3.2.8) the limits for the Fourier analysis.
Fourier analysis of the a.c. current waveform, including the effect of com-
mutation (Fig. 3.4) leads to the following relationship between the rms of the
fundamental component and the direct current"):

where
,/ [cos 2a - cos 2(a + u)I2 + [2u + sin 2a - sin 2(a + u)]
k=
~ [ C O Sa - cos(a + u)]

for values of u not exceeding 60".

 The values of k are very close to unity under normal operating conditions, i.e.,
when the voltage and currents are close to their nominal values and the a.c.
voltage waveforms are symmetrical and undistorted. Alternative steady state
models for operating conditions deviating from the above are described in
Chapters 7 and 10.
Taking into account the transformer tap position the current on the primary
side becomes

When using per unit values based on a common power and voltage base on both
sides of the converter, the direct current base has to be fi
times larger than the
a.c. current base (as explained in Section 6.3) and equation (3.2.10) becomes

Using the fundamental components of voltage and current and assuming perfect
filtering at the converter terminals the power factor angle at the converter
terminals is 4 (the displacement between fundamental voltage and current
waveforms) and we may write

and

3.3. INVERSION
Owing to the unidirectional nature of current flow through the converter valves,
power reversal (i.e. power flow from the d.c. to the a.c. side) requires direct voltage
polarity reversal. This is achieved by delay angle control, which, in the absence of
commutation overlap produces rectification between 0" < a < 90" and inversion
between 90" < a < 180". In the presence of overlap, the value of 'a' at which
inversion begins is always less than 90". Moreover, unlike with rectification, full
inversion (i.e. a = 180")can not be achieved in practice. This is due to the existence
of a certain deionization angle y at the end of the conducting period, before the
voltage across the commutating valve reverses, i.e.

If the above condition is not met (yo being the minimum required extinction
angle) a commutation failure occurs; this event would upset the normal
conducting sequence and preclude the use of the steady-state model derived in
this chapter.
The inverter voltage, although of opposite polarity with respect to the rectifier,
is usually expressed as positive when considered in isolation.

Fig. 3.5. Invertor waveforms. (a) Alternating cur-

rent; (b) D.c. voltage waveforms

Typical inverter voltage and current waveforms are illustrated in Fig. 3.5. By
similarity with the waveforms of Fig. 3.4, the following expression can be written
for the inverter voltage in terms of the extinction angle:

which is the same as equation (3.2.5) substituting y for a.

It should by now be obvious that inverter operation requires the existence of
three conditions as follows:
6) An active a.c. system which provides the commutating voltages.
(ii) A d.c. power supply of opposite polarity to provide continuity for the
unidirectional current flow (i.e. from anode to cathode through the
switching devices).
(iii) Fully controlled rectification to provide firing delays beyond 90".
When these three conditions are met, a negative voltage of a magnitude given by
equation (3.3.1), is impressed across the converter bridge and power ( - VdId)is
inverted. Note that the power factor angle (4) is now larger than 90°, i.e.
P = , , ~ v , ~ ~cos
, I ,4 = - fi~,~
cos ,
(n~
- 4), (3.3.2)

,~, ,Q,~~=
,~i f ,~, ,4,~~=
sin ,~i f sin (n - 4 ) (3.3.3)
 Fig. 3.6. P and Q vector diagram

Equations (3.3.2) and (3.3.3) indicate that the inversion process still requires
reactive power supply from the a.c. side. The vector diagram of Fig. 3.6 illustrates
the sign of P and Q for rectification and inversion.

3.4. COMMUTATION REACTANCE

Figure 3.7 shows the general case of n bridges connected in parallel on the a.c.
side. In the absence of filters the pure sinusoidal voltages exist only behind the
system source impedance (X,,) and the commutation reactance (Xcj)for the jth
bridge is thus
Xcj= X,, + Xtj (3.4.1)
However, if the bridges are under the same controller or under identical
controllers then it is preferable to create a single equivalent bridge. The
commutation reactance of such an equivalent bridge depends upon the d.c.
connections and also the phase shifting between bridges.
If there are k bridges with the same phase shift then they will commutate at the
same time and the equivalent reactance must reflect this. For a series connection
of bridges the commutation reactance of the equivalent bridge is:

Fig. 3.7. 'n' bridges connected in series on the d.c. side and in
parallel on the a.c. side
where j represents any of the n bridges. For bridges connected in parallel on the
d.c. side the equivalent bridge commutation reactance is:

It should be noted that with perfect filtering or when many bridges are used
with different transformer phase shifts the voltage on the a.c. side of the converter
transformers may be assumed to be sinusoidal and hence Xs, has no influence on
the commutation.
Moreover, the presence of local plant components at the converter terminals
may affect the commutation reactance. By way of example, let us consider the two
ends of the New Zealand h.v.d.c. link (with reference to Fig. 3.8). It must be noted
that h.v.d.c. schemes are normally designed for twelve-pulse operation and that
filters are always provided (i.e. the system impedance can be ignored).

AC Benrnore Haywards
220 kV

To South
I
1 Pole I

t
To North
Sea + Islandsystem

Synchronous
cornpensotors

Fig. 3.8. Simplified diagram of the New Zealand h.v.d.c. interconnection

(i) At Haywards the effect of the subtransient reactance of the synchronous

compensators on the tertiaries of the converter transformers must be taken
into account. The approximate equivalent circuit is illustrated in Fig. 3.9
and the commutation reactance is

where
X,-transformer secondary leakage reactance
X,-transformer primary leakage reactance
X,-transformer tertiary leakage reactance
X&' is the subtransient reactance of the synchroncus condenser unit
 Fig. 3.9. Equivalent circuit for the calculation of the com-
mutation reactance at Haywards end

1
A. C.
t+ 4 system
I

Fig. 3.10. Equivalent circuit for the calculation of the commutation

reactance at the Benmore end

(ii) At the Benmore end the subtransient reactance of the generators is

combined in parallel with the secondary reactance of the interconnecting
transformer. (The primary reactance is beyond the filters and can thus be
neglected.) The approximate equivalent circuit is illustrated in Fig. 3.10.
Although there are two converter groups commutating on this reactance,
the coclmutations are not simultaneous due to the 30-degree phase-shift of
their respective transformers. Thus the effective commutation reactance
per group is

where
X-two-winding transformer leakage reactance
 X,-interconnecting transformer secondary leakage reactance (note
filters connected to tertiary winding)
X:-generator subtransient reactance
n- number of generators connected

3.5. D.C. TRANSMISSION

The sending and receiving ends of a two-terminal d.c. transmission link such as
that illustrated in Fig. 3.8 can be modelled as single equivalent bridges with
terminal voltages Vdr and Vd, respectively. The direct current is thus given by:

where Rd is the resistance of the link and includes the loop transmission resistance
(if any), the resistance of the smoothing reactors and the converter valves.
Convertor I Convertor I1

Fig. 3.1 1 . Two-terminal d.c. link

The prime considerations in the operation of a d.c. transmission system are to

minimize the need for reactive power at the terminals and reduce system losses.
These objectives require maintaining the highest possible transmission voltage
and this is achieved by minimizing the inverter end extinction angle, i.e. operating
the inverter on constant extinction angle (E.A.) control while controlling the
direct current at the rectifier end by means of temporary delay angle backed by
transformer tap-change control.
E.A. control applied to the inverter automatically varies the firing angle of
advance to maintain the extinction angle y at a constant value. Deionization
imposes a definite minimum limit on y, and the E.A. control usually maintains it
at this limit.
Constant-current (C.C.) control applied to the rectifier regulates the firing
angle a to maintain a prespecified link current I?, within the range of a. If the
value of a required to maintain I",Pfalls below its minimum limit, current control
is transferred to the inverter, i.e. a is fixed on its minimum limit, and the inverter
firing angle is advanced to control the current.
The converter-transformer tap-change is a composite part of this control. The
rectifier transformer attempts to maintain a within its permitted range. The
inverter transformer attempts to regulate the d.c. voltage at some point along the
line to a specified level. For minimum loss and minimum-reactive-power
absorption, this voltage is required to be as high as possible, and the firing angle
of the rectifier should be as low as possible.
- Rectifier ( a m i , )
---- - - ----_ -
--- ------ ---,,.
--- -- _ - b
1
lnvertor e.a

lnvertor c.c., I

- C

Fig. 3.12. Normal control characteristics

Fig. 3.13. Control characteristics and power flow

reversal
 Fig. 3.12 shows the d.c. voltage/current characteristics at the rectifier and
inverter ends (the latter have been drawn with reverse polarity in order to
illustrate the operating point). The current controller gains are very large and for
all practical purposes the slopes of the constant current characteristics can be
ignored. Consequently the operating current is equal to the relevant current
setting, i.e. IdSr and IdSifor rectifier and inverter constant current control
respectively.
The direction of power flow is determined by the current settings, the rectifier
end always having the larger setting. The difference between the settings is the
current margin Idmand is given by
Idm = Id., - Idsi > 0 (3.5.4)
Many d.c. transmission schemes are bidirectional, i.e. each converter operates
sometimes as a rectifier and sometimes as an inverter. Moreover, during d.c. line
faults, both converters are forced into the inverter mode in order to de-energize
the line faster. In such cases each converter is provided with a combined
characteristic as shown in Fig. 3.13 which includes natural rectification,constant
current control and constant extinction angle control.
With the characteristics shown by solid lines (i.e. operating at point A), power is
transmitted from Converter I to Converter 11. Both stations are given the same
current command but the current margin setting is subtracted at the inverter end.
When power reversal is to be implemented the current settings are reversed and
the broken line characteristics apply. This results in operating point B, with direct
voltage reversed and no change in direct current.

Alternative forms of control

A commonly used operating mode is constant power (C.P.) control. As with
constant current control either converter can control power. The power setting at
the rectifier terminal PdSrmust be larger than that at the invertor terminal PdSiby a
suitable power margin Pdm,that is:
Pd, = Pd,, - PdSi> 0 (3.5.5)
The C.P. controller adjusts the C.C. control setting I:p to maintain a specified
power flow P : through the link, which is usually more practical than C.C. control
from a system-operation point of view. The voltage/current loci now become
non-linear, as shown in Fig. 3.14.
Several limits are added to the C.P. characteristics as shown in Fig. 3.15. These
are:
- A maximum current limit with the purpose of preventing thermal damage to
the converter valves; normally between 1 and 1.2 times the nominal current.
- A minimum current limit (about 10 per cent of the nominal value) in order to
avoid possible current discontinuities which can cause overvoltages.
- Voltage dependent current limit (line OA in the figure) in order to reduce the
power loss and reactive power demand.
 Fig. 3.14. Constant power characteristics

Fig. 3.15. Voltage and current limits

In cases where the power rating of the d.c. link is comparable with the rating of
either the sending or receiving a.c. systems interconnected by the link, the
frequency of the smaller a.c. system is often controlled to a large extent by the d.c.
link. With power-frequency (P.F.) control if the frequency goes out of pre-
specified limits, the output power is made proportional to the deviation of
frequency from its nominal value. Frequency control is analogous to the current
control described earlier, i.e. the converter with lower voltage determines the
direct voltage of the line and the one with higher voltage determines the
frequency. Again, current limits have to be imposed, which override the frequency
error signal.
C.P.1E.A. and C.C.1E.A. controls were evolved principally for bulk point-to-
point power transmission over long distances or submarine crossings and are still
the main control modes in present use.
 Multiterminal d.c. schemes are also being considered, based on the basic
controls already described. Two alternatives are possible, i.e. constant voltage
parallel"' and constant current series(@schemes.

3.6. REFERENCES
1. C. Adamson and N. G. Hingorani, 1960.H.VD.C. Power Transmission. Garraway Ltd.,
London.
2. E. W. Kimbark, 1971. Direct Current Transmission.Vol. I . Wiley Interscience, London.
3. B. J. Cory (ed.), 1965. High-Voltage Direct-Current Converters and Systems. Macdonald,
London.
4. E. Uhlmann, 1975. Power Transmission by Direct Current. Springer-Verlag,
Berlin/Heidelberg.
5. U. Lamm and E. Uhlmann and P. Danfors, 1963. 'Some aspects of tapping of h.v.d.c.
transmission systems', Direct Current, 8, 124-1 29.
6. C. Adamson and J. Arrillaga, 1968. 'behaviour of multiterminal a.c.4.c. intercon-
nections with series-connected stations', Proc. LEE, 115 (1 l), 1685-1692.
 Load Flow
4.1. INTRODUCTION
Under normal conditions electrical transmission systems operate in their steady-
state mode and the basic calculation required to determine the characteristics of
this state is termed load flow (or power flow).
The object of load-flow calculations is to determine the steady-state operating
characteristics of the power generation/transmission system for a given set of
busbar loads. Active power generation is normally specified according to
economic-dispatching practice and the generator voltage magnitude is normally
maintained at a specified level by the automatic voltage regulator acting on the
machine excitation. Loads are normally specified by their constant active and
reactive power requirement, assumed unaffected by the small variations of
voltage and frequency expected during normal steady-state operation.
The solution is expected to provide information of voltage magnitudes and
angles, active and reactive power flows in the individual transmission units, losses
and the reactive power generated or absorbed at voltage-controlled buses.
The load flow problem is formulated in its basic analytical form in this chapter
with the network represented by linear, bilateral and balanced lumped para-
meters. However the power and voltage constraints make the problem non-linear
and the numerical solution must therefore be iterative in nature.
The currents problem faced in the development of load flow are: an ever
increasing size of systems to be solved, on-line applications for automatic control,
and system optimization. Hundreds of contributions have been offered in the
literature to overcome these problems.(')
Five main properties are required of a load-flow solution method.
(i) High computational speed. This is especially important when dealing with
large systems, real time applications (on-line), multiple case load-flow
such as in system security assessment, and also in inter-active
applications.
(ii) Low computer storage. This is important for large systems and in the use
of computers with small core storage availability, e.g. mini-computers for
on-line application.
(iii) Reliability of solution. It is necessary that a solution be obtained for
 illconditioned problems, in outage studies and for real time applications.
(iv) Versatility. An ability on the part of load-flow to handle conventional and
special features (e.g. the adjustment of tap ratios on transformers; different
representations of power-system apparatus), and its suitability for
incorporation into more complicated processes.
(v) Simplicity. The ease of coding a computer program of the load-flow
algorithm.
The type of solution required from a load-flow also determines the method
used :
accurate or approximate
unadjusted or adjusted
off-line or on-line
single case or multiple cases
The first column are requirements needed for considering optimal load-flow and
stability studies, and the second column those needed for assessing security of a
system. Obviously, solutions may have a mixture of the properties from either
column.
The first practical digital solution methods for load flow were the Y matrix-
iterative methods.(2)These were suitable because of the low storage requirements,
but had the disadvantage of converging slowly or not at all. Z matrix methodd3)
were developed which overcame the reliability problem but a sacrifice was made
of storage and speed with large systems.
The Newton-Raphson m e t h ~ d ( ~ )was
. ( ~developed
) at this time and was found
to have very strong convergence. It was not, however, made competitive until
sparsity programming and optimally ordered Gaus~ian-elimination(~)~(~)~(*) were
introduced, which reduced both storage and solution time.
Nonlinear programming and hybrid methods have also been developed, but
these have created only academic interest and have not been accepted by
industrial users of load flow. The Newton-Raphson method and techniques
derived from this algorithm, satisfy the requirements of solution type and
programming properties better than previously used techniques and are
gradually replacing them.

4.2. BASIC NODAL METHOD

In the nodal method as applied to power-system networks, the variables are the
complex node (busbar) voltages and currents, for which some reference must be
designated. In fact, two different references are normally chosen: for voltage
magnitudes the reference is ground, and for voltage angles the reference is chosen
as one of the busbar voltage angles, which is fixed at the value zero (usually). A
nodal current is the nett current entering (injected into) the network at a given
node, from a source and/or load external to the network. From this definition, a
current entering the network (from a source) is positive in sign, while a current
leaving the network (to a load) is negative, and the nett nodal injected current is
the algebraic sum of these. One may also speak in the same way of nodal injected
powers S = P +jQ.

.,
Fig. 4.1. Simple network showing nodal quantities

Figure 4.1 gives a simple network showing the nodal currents, voltages and
powers.
In the nodal method, it is convenient to use branch admittances rather than
impedances. Denoting the voltages of nodes k and i as Ekand Ei respectively, and
the admittance of the branch between them as yki,then the current flowing in this
branch from node k to node i is given by:

Let the nodes in the network be numbered 0, 1 . . .n, where 0 designates the
reference node (ground). By Kirchhoffs current law, the injected current I, must
be equal to the sum of the currents leaving node k, hence:

Since Eo = 0, and if the system is linear,

If this equation is written for all the nodes except the reference, i.e. for all busbar
in the case of a power-system network, then a complete set of equations defining
the network is obtained in matrix form as:
where
n

Ykk= C yki = self-admittance of node k

i=O#k

Y,, = - yki = mutual-admittance between nodes k and i

In shorthand matrix notation, equation (4.2.4) is simply

or in summation notation, we have

The nodal admittance matrix in equations (4.2.4) or (4.2.5) has a well-defined

structure, which makes it easy to construct automatically. Its properties are as
follows :
(i) Square of order n x n
(ii) Symmetrical, since yki = yik
(iii) Complex
(iv) Each off-diagonal element yki is the negative of the branch admittance
between nodes k and i, and is frequently of value zero.
(v) Each diagonal element ykk is the sum of the admittance of the branches
which terminate on node k, including branches to ground.
(vi) Because in all but the smallest practical networks very few nonzero
mutual admittances exist, matrix Y is highly sparse.

4.3. CONDITIONING OF Y MATRIX

The set of equations I = Y .E may or may not have a solution. If not, a simple
physical explanation exists, concerning the formulation of the network problem.
Any numerical attempt to solve such equations is found to break down at some
stage of the process (what happens in practice is usually that a finite number is
divided by zero).
The commonest case of this is illustrated in the example of Fig. 4.2.

Fig. 4.2. Example of singular network

 The nodal equations are constructed in the usual way as:

Suppose that the injected currents are known, and nodal voltages are
unknown. In this case no solution for the latter is possible. The Y matrix is
described as being singular, i.e. it has no inverse, and this is easily detected in this
example by noting that the sum of the elements in each row and column is zero,
which is a sufficient condition for singularity, mathematically speaking. Hence, if
it is not possible to express the voltages in the form E = Y -'. I, it is clearly
impossible to solve equation (4.3.1.)by any method, whether involving inversion
of Y or otherwise.
The reason for this is obvious: we are attempting to solve a network whose
reference node is disconnected from the remainder, i.e. there is no effective
reference node, and an infinite number of voltage solutions will satisfy the given
injected current values.
When, however, a shunt admittance from at least one of the busbar in the
network of Fig. 4.2. is present, the problem of insolubility immediately vanishes in
theory, but not necessarily in practice. Practical computation cannot be
performed with absolute accuracy, and during a sequence of arithmetic
operations, rounding errors due to working with a finite number of decimal
places accumulate. If the problem is well-conditioned and the numerical solution
technique is suitable, these errors remain small and do not mask the eventual
results. If the problem is ill-conditioned, and this usually depends upon the
properties of the system being analysed, any computational errors introduced are
likely to become large with respect to the true solution.
It is easy to see intuitively that if a network having zero shunt admittances
cannot be solved even when working with absolute computational accuracy, then
a network having very small shunt admittances may well present dficulties when
working with limited computational accuracy. This reasoning provides a key to
the practical problems of network, i.e. Y matrix, conditioning. A network with
shunt admittances which are small with respect to the other branch admittances
is likely to be ill-conditioned, and the conditioning tends to improve with the size
of the shunt admittances, i.e. with the electrical connection between the network
busbars and the reference node.

4.4. THE CASE WHERE ONE VOLTAGE IS KNOWN

In load-flow studies, it usually happens that one of the voltages in the network is
specified, and instead the current at that busbar is unknown. This immediately
 63
alleviates the problem of needing at least one good connection with ground,
because the fixed busbar voltage can be interpreted as an infinitely strong ground
tie. If it is represented as a voltage source with a series impedance of zero value,
and then converted to the Norton equivalent, the fictitious shunt admittance is
infinite, as is the injected current. This approach is not computationally feasible,
however.
The usual way to deal with a voltage which is fixed is to eliminate it as a
variable from the nodal equations. Purely for the sake of analytical convenience,
let this busbar be numbered 1 in an n-busbar network. The nodal equations are
then
I l = Y E + Y E + ..............
11 1 12 2 Y l ,En
I 2 = Y 2 , E 1 + Y E + ..............
22 2 Y2nEn
(4.4.1)
In = Y n l E l + Y n , E 2 + ..............YnnEn
The terms in El on the right-hand side of equations (4.4.1) are known
quantities, and as such are transferred to the left-hand side.
1 , - Y E = Y E + ..............
11 1 12 2 Y l ,En
I 2 - Y 2 , E 1 = Y E + ..............
22 2 Y2nEn
(4.4.2)
In - Y n l E l = Y n 2 E 2+ .............. YnnEn

The first row of this set may now be eliminated, leaving (n - 1)equations in (n - 1)
unknowns, E 2 . .. En. In matrix form, this becomes:

The new matrix Y is obtained from the full admittance matrix Y merely by
removing the row and column corresponding to the fixed-voltage busbar, both in
the present case where it is numbered 1 or in general.
In summation notation, the new equations are
n
1,- y , , ~ =
, 1 YkiEi for k = 2 ...n (4.4.5)
i=2

which is an (n - 1) set in (n - 1) unknowns. The equations.are then solved by any

of the available techniques for the unknown voltages. It is noted that the problem
of singularity when there are no ground ties disappears if one row and column are
removed from the original Y matrix.
Eliminating the unknown current I , and the equation in which it appears is the
simplest way of dealing with the problem, and reduces the order of the equations
by one. I , is evaluated after the solution from the first equation in equation (4.4.1).

4.5. ANALYTICAL DEFINITION OF THE PROBLEM

The complete definition of power flow requires knowledge of four variables at
each bus k in the system.
P,-real or active power
Q,-reactive or quadrature power
Vk-voltage magnitude
0,-voltage phase angle
only two are known apriori to solve the problem, and the aim ofthe load flow is to
solve the remaining two variables at a bus.
We define three different bus conditions based on the steady-state assumptions
of constant system frequency and constant voltages, where these are controlled.
(i) Voltage controlled bus. The total injected active power P, is specified, and
the voltage magnitude Vk is maintained at a specified value by reactive
power injection. This type of bus generally corresponds to a generator
where P, is fixed by turbine governor setting and V, is fixed by automatic
voltage regulators acting on the machine excitation; or a bus where the
voltage is fixed by supplying reactive power from static shunt capacitors
or rotating synchronous compensators, e.g. at substations.
+
(ii) Nonvoltage-controlled bus. The total injected power, P, jQ,, is specified
at this bus. In the physical power system this corresponds to a load centre
such as a city or an industry, where the consumer demands his power
requirements. Both P, and Qk are assumed to be unaffected by small
variations in bus voltage.
(iii) Slack (swing)bus. This bus arises because the system losses are not known
precisely in advance of the load flow calculation. Therefore the total
injected power cannot be specified at every single bus. It is usual to choose
one of the available voltage controlled buses as slack, and to regard its
active power as unknown. The slack bus voltage is usually assigned as the
system phase reference, and its complex voltage
Es = Vs&
is therefore specified. The analogy in a practical power system is the
generating station which has the responsibility of system frequency
control.
Load flow solves a set of simultaneous nonlinear algebraic power equations for
 INPUT
Read system data and the
specified loads and gen-
eration. Also read the
voltage specifications at

and angles at all system

T IUpdate voltages and

angles in order to
1
Iteration
satisfy the specified
cycle
conditions of load and
1 generation I

generation and all line

I power flows. losses etc. I
Fig. 4.3. Flow diagram of basic load-flow algorithm

the two unknown variables at each node in a system. A second set of variable
equations, which are linear, are derived from the first set, and an iteration method
is applied to this second set.
The basic algorithm which load-flow programs use is depicted in Fig. 4.3.
System data, such as busbar power conditions, network connections and
impedance, are read in and the admittance matrix formed. Initial voltages are
+
specified to all buses; for base case load flows, P, Q buses are set to 1 j O while
P, V busbars are set to V + jO.
The iteration cycle is terminated when the busbar voltages and angles are such
that the specified conditions of load and generation are satisfied. This condition is
accepted when power mismatches for all buses are less than a small tolerance, q,,
or voltage increments less than q2. Typical figures for q, and q2 are 0.01 p.u. and
0.001 p.u. respectively. The sum of the square of the absolute values of power
mismatches is a further criterion sometimes used.
When a solution has been reached, complete terminal c lditions for all buses are
computed. Line power flows and losses and system totals can then be calculated.
66
4.6. NEWTON-RAPHSON METHOD OF SOLVING LOAD mows
The generalized Newton-Raphson method is an iterative algorithm for solving a
set of simultaneous nonlinear equations in an equal number of unknowns.
f,(x,)=O for k = 1 + N (4.6.1)
m=l+N
At each iteration of the N - R method, the nonlinear problem is approximated
by the linear matrix equation. The linearizing approximation can best be
visualized in the case of a single-variable problem.

Fig. 4.4. Singlc-variable linear approximately

In Fig. 4.4. xPis an approximation to the solution, with error AxP at iteration p.
Then

This equation can be expanded by Taylor's theorem,

If the initial estimates of the variable xP is near the solution value, AxP will be
relatively small and all terms of higher powers can be neglected. Hence,
+
f ( x p ) AxP f ' ( x P =
) 0 (4.6.4)

The new value of the variable is then obtained from

Xp+l = xp + AXP
Equation (4.6.4)may be rewritten as
f ( x P )= - JAxP
 The method is readily extended to the set of N equations in N unknowns. J
becomes the square Jacobian matrix of first order partial differentials of the
functions fk(xm).Elements of [ J ] are defined by

and represents the slopes of the tangent hyperplanes which approximate the
functions fk(xm)at each iteration point.
The Newton-Raphson algorithm will converge quadratically if the functions
have continuous first derivates in the neighbourhood of the solution, the
Jacobian matrix is nonsingular, and the initial approximations of x are close to
the actual solutions. However the method is sensitive to the behaviours of the
functions fk(xm)and hence to their formulation. The more linear they are, the
more rapidly and reliably Newton's method converges. Nonsmoothness, i.e.
humps, in any one of the functions in the region of interest, can cause convergence
delays, total failure or misdirection to a nonuseful solution.

Equations relating to power-system load flow

The nonlinear network governing equations are

I,= C ykmEm for all k

me k

where I, is the current injected into a bus k. The power at a bus is then given by

Mathematically speaking, the complex load-flow equations are nonanalytic, and

cannot be differentiated in complex form. In order to apply Newton's method,
the problem is separated into real equations and variables. Polar or rectan-
gular coordinates may be used for the bus voltages. Hence we obtain two
equations

In polar coordinates the real and imaginary parts of equation (4.6.10) are

Pk= x
me k
VkVm(Gkm
cos 8 , +B, sin 8,,) (4.6.11)

Qk= VkVm(Gh sin 8 , - B , cos 8,) (4.6.12)

me k
where

Linear relationships are obtained for small variations in the variables 0 and V by
forming the total differentials, the resulting equations being:
For a PQ busbar:

APk = c ap
~
m s k aOm
0 + ,LA
ap
1 vm
m s k avm
and

For a PV busbar:
Only equation (4.6.13)is used, since Qk is not specified.
For a slack busbar:
No equations.
The voltage magnitudes appearing in equations (4.6.13)and (4.6.14)for PV and
slack busbars are not variables, but are fixed at their specified values. Similarly 0
at the slack busbar is fixed.
The complete set of defining equations is made up of two for each PQ busbar
and one for each PV busbar. The problem variables are V and 0 for each PQ
busbar and 0 for each PV busbar. The number of variables is therefore equal to
the number of equations. Algorithm (4.6.7) then becomes:

P mismatches 9 corrections
for all P Q for all PQ and
and PV busbars PV busbars
Q mismatches AVp V corrections
for all PQ busbars

Jacobian matrix (4.6.15)

The division of each AVP by Vf -'does not numerically affect the algorithm, but
simplifies some of the Jacobian-matrix terms. For busbars k and rn (not row k and
column rn in the matrix)
apk = VkVm(Gkm
Hkm= -- sin Ohm - B , cos 0,)
a0m
apk
cos ohm+ Bkmsin Ohm)
Nkm= Vm-= VkVm(Gkm
av m

aQk
Lkm= Vm- = VkVm(Gkm
sin 0 , - B, cos Ohm)
av m
and for m = k:

In practice, some programs express these coefficients using voltages in rec-

+
tangular form, i.e. ei j f i . This only effects the speed of calculation of the
mismatches and the matrix elements, by eliminating the time-consuming
trigonometrical functions.
In rectangular coordinates the complex power equations are given as

and these are divided into real and imaginary parts

At a voltage controlled bus the voltage magnitude is fixed but not the phase angle.
Hence both ek and f k vary at each iteration. It is necessary to provide another
equation
V,"=e,t+f,Z
to be solved with the real power equation for these buses.
Linear relationships are obtained for small variations in e andf , by forming the
total differentials,
apk
AP,= C ---At?,+ C -apk
Afm
m s k aem msk ' f m

for all buses except the slack bus;

= C
msk
U k m Aem +C
msk
Wkm A f m
70
for all nonvoltage controlled buses;

for voltage controlled buses.

The Jacobian matrix has the form

and the values of the partial differentials, which are the Jacobian elements, are
given by

Skk + Gkkek + Bkkfk

= ak

Wkk = a, - G k k e k - B k k f k

Tkk = bk - B k k e k +Gkkfk
Ukk = - bk - B k k e k +Gkkfk
E E , = 2ek
FFk = 2 f k

For voltage controlled buses, V is specified, but not the real and imaginary
components of voltage, e and f . Approximations can be made, for example, by
ignoring the off-diagonal elements in the Jacobian matrix, as the diagonal
elements are the largest. Alternatively for the calculation of the elements the
voltages can be considered as E = 1 +jO. The off-diagonal elements then become
constant.
The polar coordinate representation appears to have computational advan-
tages over rectangular coordinates. Real power mismatch equations are present
for all buses except the slack bus, while reactive power mismatch equations are
needed for nonvoltage controlled buses only.
The Jacobian matrix has the sparsity of the admittance matrix [Y] and has
positional but not numerical symmetry. To gain in computation, the form of
[A& AV/ VIT is normally used for the variable voltage vector. Both increments
are dimensionless and the Jacobian coefficients are now symmetric in structure
though not in value. The values of [J] are all functions of the voltage variables V
and 8 and must be recalculated each iteration.
As an example, the Jacobian matrix equation for the four-busbar system of
Fig. 4.5 is given as equation (4.6.17)

QSlack
1,

I
Fig. 4.5. Sample system

The differences in bus powers are obtained from

AP, =P;rP -pk
AQk = QfP - Qk
A further improvement is to replace the reactive power residual AQ in the
Jacobian matrix equations by AQ/ V. The performance of the Newton-Raphson
method is closely associated with the degree of problem non-linearity; the best
left-hand defining functions are the most linear ones. If the system power
equation (4.6.19)is divided throughout by Vk,only one term Q",/ Vk on the right-
hand side of this equation is nonlinear in Vk.For practical values of Q",Pand Vk,
this nonlinear term is numerically relatively small. Hence it is preferable to use
AQ/V instead of AQ in the Jacobian matrix equation.
Dividing AP by V is also helpful, but is less effective since the real power
component of the problem is not strongly coupled with voltage magnitudes. A
further alternative is to formulate current residuals at a bus. While com-
putationally simple, this method shows poor convergence in the same way as Y
matrix iterative methods.
A flow diagram of the basic Newton-Raphson algori~hmis given in Fig. 4.6.
 -
INPUT DATA
Read all input data ,form
system admittance matrix,
initialize voltages and angles
at all busbars

AForm Jaoobian matrix

for voltage and angle

results

Fig. 4.6. Flow diagram of the basic

Newton-Raphson load-flow algorithm

4.7. TECHNIQUES WHICH MAKE THE NEWTON-RAPHSON

METHOD COMPETITIVE IN LOAD n o w
The efficient solution of (4.6.15) at each iteration is crucial to the success of the
N-R method. If conventional matrix techniques were to be used, the storage (an2)
and computing time (an3) would be prohibitive for large systems.
For most power-system networks the admittance matrix is relatively sparse,
and in the Newton-Raphson method of load-flow the Jacobian matrix has this
same sparsity.
The techniques which have been used to make the Newton-Raphson
competitive with other load flow methods involve the solution of the Jacobian
matrix equation and the preservation of the sparsity of the matrix by ordered
triangular factorization.

Sparsity programming
In conventional matrix programming, double subscript arrays are used for the
location of elements. With sparsity programming,(6)only the nonzero elements
are stored, in one or more vectors, plus integer vectors for identification.
For the admittance matrix of order n the conventional storage requirements
are nZwords, but by sparsity programming 66 + 3n words are required, where b is
the number of branches in the system. Typically b = lSn, and the total storage is
12n words. For a large system (say 500 buses) the ratio of storage requirements of
conventional and sparse techniques is about 40: 1.

Triangular factorization
To solve the Jacobian matrix equation (4.6.15), represented here as

for increments in voltage, the direct method is to find the inverse of [JI and solve
for [AEJ from
CAE3 = [A - [Asl' (4.7.1)
In power systems [J] is usually sparse but [JI-' is a full matrix.
The method of triangular factorization solves for the vector [AEJ by
eliminating [J] to an upper triangular matrix with a leading diagonal, and then
back-substituting for [ A q , i.e. eliminate to
CAS'l =C u l CAE3
and back-substitute
[U] - [AS'] = [AEI

The triangulation of the Jacobian is best done by rows. Those rows below the
one being operated on need not be entered until required. This means that the
maximum storage is that of the resultant upper triangle and diagonal. The lower
triangle can then be used to record operations.
The number of multiplications and additions to triangulate a full matrix is $N3,
compared to N3 to find the inverse. With sparsity programming the number of
operations varies as a factor of N. If rows are normalized N further operations are
saved.

Optimal ordering
In power-system load flow, the Jacobian matrix is usually diagonally dominant
which implies small round-off errors in computation. When a sparse matrix is
triangulated, nonzero terms are added in the upper triangle. The number added
is affected by the order of the row eliminations, and total computation time
increases with more terms.
The pivot element is selected to minimize the accumulation of nonzero terms,
and hence conserve sparsity, rather than minimizing round off error. The
diagonals are used as pivots.
Optimal ordering of row eliminations to con ,sparsity is a practical
impossibility due to the complexity of programming m d time involved. However,
semioptimal schemes are used and these can be divided into two sections.
(a) Pre~rdering.'~'
Nodes are renumbered before triangulation. No complicated
programming or storage is required to keep track of row and column
interchanges.
(i) Nodes are numbered in sequence of increasing number of connected lines.
(ii) Diagonal banding-nonzero elements are arranged about either the major
or minor diagonals of the matrix.
(b) Dynamic ordering.") Ordering is effected at each row during the elimination.
(i) At each step in the elimination, the next row to be operated on is that with
the fewest nonzero terms.
(ii) At each step in the elimination, the next row to be operated on is that
which introduces the fewest new nonzero terms, one step ahead.
(iii) At each step in the elimination, the next row to be operated on is that
which introduces the fewest new nonzero terms, two steps ahead. This may
be extended to the fully optimal case of looking at the effect in the final
step.
(iv) With cluster ordering, the network is subdivided into groups which are
then optimally ordered. This is most efficient if the groups have a
minimum of physical intertie. The matrix is then anchor banded.
The best method arises from a trade-off between a processing sequence which
requires the least number of operations, and time and memory requirements.
The dynamic ordering scheme of choosing the next row to be eliminated as that
with the fewest nonzero terms, appears to be better than all other schemes in
sparsity conservation, number of arithmetic operations required, ordering times
and total solution time.
However, there are conditions under which other ordering would be prefer-
able, e.g. with system changes affecting only a few rows these rows should be
numbered last; when the subnetworks have relatively few interconnections it is
better to use cluster ordering.

Aids to convergence
The N-R method can diverge very rapidly or converge to the wrong solution if
the equations are not well behaved or if the starting voltages are badly chosen.
Such problems can often be overcome by a variety of techniques. The simplest
device is to impose a limit on the size of each A8 and AV correction at each
iteration. Figure 4.7. illustrates a case which would diverge without this device.
Another more complicated method is to calculate good starting values for the
8's and V's, which also reduces the number of iterations required.
In power-system load flow, setting voltage controlled buses to V+jO and
nonvoltage controlled buses to 1 + j O may give a poor starting point for the
Newton-Raphson method.
If previously stored solutions for a network are available these should be used.
One or two iterations of a Y matrix iterative method(2)can be applied before
 Fig. 4.7. Example of diverging solution

commencing the Newton method. This shows fast initial convergence unless the
problem is ill-conditioned, in which case divergence occurs.
A more reliable method is the use of one iteration of a d.c. load flow (i.e.
neglecting losses and reactive power conditions) to provide estimates of voltage
angles; followed by one iteration of a similar type of direct solution to obtain
voltage magnitudes. The total computing time for both sets of equations is about
50 percent of one Newton-Raphson iteration and the extra storage required is
only in the programming statements. The resulting combined algorithm is faster
and more reliable than the formal Newton method and can be used to monitor
diverging or difficult cases, before commencing the Newton-Raphson algorithm.

4.8. CHARACTERISTICS OF THE NEWTON-RAPHSON LOAD FLOW

With sparse programming techniques and optimally ordered triangular factori-
zation, the Newton method for solving load flow has become faster than other
methods for large systems. The number of iterations is virtually independent of
system size (from a flat voltage start and with no automatic adjustments) due to
the quadratic characteristic of convergence. Most systems are solved in 2 + 5
iterations with no acceleration factors being necessary.
With good programming the time per iteration rises nearly linearly with the
number of system buses N, so that the overall solution time varies as N. One
Newton iteration is equivalent to about seven Gauss-Seidel iterations. For a 500-
bus system, the conventional Gauss-Seidel method takes about 500 iterations
and the speed advantage of the Newton method is then 15:l. Storage
requirements of the Newton method are greater, however, but increase linearly
with system size. It is therefore attractive for large systems.
The Newton method is very reliable in system solving, given good starting
approximations. Heavily loaded systems with phase shifts up to 90" can be
solved. The method is not troubled by ill-conditioned systems and the location of
slack bus is not critical.
 Due to the quadratic convergence of bus voltages, high accuracy (near exact
solution) is obtained in only a few iterations. This is important for the use of load
flow in short-circuit and stability studies. The method is readily extended to
include tap-changing transformers, variable constraints on bus voltages, and
ieactive and optimal power scheduling. Network modifications are easily made.
The success of the Newton method is critical on the formulation of the
problem-defining equations. Power mismatch representation is better than the
current mismatch versions. To help negotiate nonlinearities in the defining
functions, limits can be imposed on the permissible size of voltage corrections at
each iteration. These should not be too small, however, as they may slow down
the convergence for well-behaved systems,
The coefficients of the Jacobian matrix are not constant, they are functions of
the voltage variables V and 8, and hence vary each iteration. However, after a few
iterations, as V and 8 tend to their final values the coefficients will tend to
constant values.
One modification to the Newton algorithm is to calculate the Jacobian for the
first two or three iterations only and then use the final one for all the following
iterations. Alternatively the Jacobian can be updated every two or more
iterations. Neither of these modifications greatly affects the convergence of the
algorithm, but much time is saved (but not storage).

4.9. DECOUPLED NEWTON LOAD FLOW

An inherent characteristic of any practical electric power transmission system
operating in the steady-state condition is the strong interdependence between
active powers and bus voltage angles, and between reactive powers and voltage
magnitudes. Correspondingly, the coupling between these 'P-8' and 'Q-V'
components of the problem is relatively weak. Many algorithms have been
proposed which adopt this decoupling prin~iple.(~)~('~)~(")
The voltage vectors method uses a series approximation for the sine terms
which appear in the system defining equations, to calculate the Jacobian elements
and arrive at two decoupled equations

where for the reference node 80 = 0 and Vk= Vo. The values of Sk and 22,
represent real and reactive power quantities respectively and [7'l and [U]are
given by
where Zkmand X,, are the branch impedance and reactance respectively. [U] is
constant valued and needs be triangulated once only for a solution. [7-'J is
recalculated and triangulated each iteration.
The two equations (4.9.1)and (4.9.2)are solved alternately until a solution is
obtained. These equations can be solved using Newton's method, by expressing
the Jacobian equations as

where
CAP] = CAP]

and T and U are therefore defined in equations (4.9.3) to (4.9.6).

The most successful decoupled load flow is that based on the Jacobian matrix
equation for the formal Newton method, i.e.

If the submatrices N and J are neglected, since they represented the weak
coupling between 'P-8' and 'Q-V', the following decoupled equations result
CAP] = [HI [A81 (4.9.11)

It has been found that the latter equation is relatively unstable at some distance
from the exact solution due to the nonlinear defining functions. An improvement
in convergence is obtained by replacing this with the polar current-mismatch
formulation(7)
CAI] = [Dl [A vl (4.9.13)
Alternatively the right-hand side of both equations (4.9.11) and (4.9.12)is divided
by voltage magnitude V
 [APIvI = [ A ] [A81 (4.9.14)

The equations are solved successively using the most up-to-date values of V and 8
available. [ A ] and [ C ] are sparse, nonsymmetric in value and are both functions
of V and 8.
They must be calculated and triangulated each iteration.
Further approximations that can be made are to assume that E , = 1.0 p.u., for
all buses, and G,, < B,, in calculating the Jacobian elements. The off-diagonal
terms then become symmetric about the leading diagonal.
The decoupled Newton method compares very favourably with the formal
Newton method. While reliability is just as high for ill-conditioned problems, the
decoupled method is simple and computationally efficient. Storage of the
Jacobian and matrix triangulation is saved by a factor of 4, or an overall saving of
30-40 percent on the formal Newton load flow. Computation time per iteration is
also less than the Newton method.
However, the convergence characteristics of the decoupled method are linear,
the quadratic characteristics of the formal Newton being sacrified. Thus, for high
accuracies, more iterations are required. This is offset for practical accuracies by
the fast initial convergence of the method. Typically, voltage magnitudes
converge to within 0.3 percent of the final solution on the first iteration and may
be used as a check for instability. Phase angles converge more slowly than voltage
magnitudes but the overall solution is reached in 2 + 5 iterations. Adjusted
solutions (the inclusion of transformer taps, phase shifters, interarea power
transfers, Q and V limits) take many more iterations.
The Newton methods can be expressed as follows(12):

where
E= 1 for the full Newton-Raphson method
E =0 for the decoupled Newton algorithm
A Taylor series expansion of the Jacobian about E = 0 results in a first-order
approximation of the Newton-Raphson method whereas the decoupled method
is a zero order approximation. The method has quadratic convergence properties
because of the coupling, but retains storage requirements similar to that of the
decoupled method.

4.10. FAST DECOUPLED LOAD FLOW

By further simplifications and assumptions, based on the physical properties o f a
practical system, the Jacobians of the decoupled Newton load flow can be made
constant in value. This means that they need be triangulated only once per
solution or for a particular network.
For ease of reference, the real and reactive power equations at a node k are
reproduced here.
Pk = Vk 1Vm(Gkm
me k
cos Okm Bkmsin Okm) + (4.10.1)

Qk = vk 1
msk
vm(Gkm sin O k m - Bkm cOs e k m ) (4.10.2)
where Okm = 8, - 8,.
A decoupled method which directly relates powers and voltages is derived
using the series approximations for the trigonometric terms in equations (4.10.1)
and (4.10.2).

The equations, over all buses, can be expressed in their simplified matrix form

where P and Q are terms of real and reactive power respectively and

'kk= 1
mok
'kmBkm

tkm-tap ratio if a transformer is in the line.

A modification suggested is to replace equation (4.10.3)by
[A] [el = [PI
where

Hence [A] becomes constant valued.

 A similar direct method is obtained from the decoupled voltage vectors
method (equations (4.9.1) and (4.9.2)). If Vm,Vk are put as 1.0 p.u. for the
calculation of matrix [TI, then [TI becomes constant and need be triangulated
once only. This same simplification can be used in the decoupled voltage vectors
and Newton's method of equations (4.9.8) and (4.9.9).
Fast decoupled load flow algorithms@)are also derived from the Jacobian
matrix equations of Newton's method (equations (4.9.10)), and the decoupled
version (equations (4.9.1 1) and (4.9.12)).
If we make the assumptions
(i) E,, Em= 1.0 p.u.
(ii) Gkm<Bkm, and hence can be ignored (for most transmission line
reactance/resistance ratios, XIR 9 1)
(iii) cos (6, - 6,) A 1.0
sin (6, - 6,) A 0.0
since angle differences across transmission lines are small under. normal
loading conditions.
This leads to the decoupled equations
[AP] = [B] [A61 of order (N - 1)
[AQ] = [B] [AV] of order (N - M)
where the elements of [B] are
Bkm= - Bkm for m # k
Bkk = C Bkm
mwk

and Bkmare the imaginary parts of the admittance matrix. To simplify still further,
line resistances may be neglected in the calculation of elements of [B].
An improvement over equations (4.10.5) and (4.10.6) is based on the decoupled
equations (4.9.14) and (4.9.15) which have less nonlinear defining functions.
Applying the same assumptions listed previously, we obtain the equations
[APIV] = [B*] [A61 (4.10.7)

A number of refinements make this method very successful:

(a) Omit from the Jacobian in equation (4.10.7) the representation of those
network elements that predominantly affect MVAR or reactive power
flow, e.g. shunt reactances and off-nominal in-phase transformer taps.
Neglect also the series resistances of lines.
(b) Omit from the Jacobian of equation (4.10.8) the angle shifting effects of
phase shifters.
The resulting fast decoupled load flow equations are then,
 CAP/ V ] = [B'] [AO]
CAQIVI = CB"1[ A vl
where
1
B;, = - -, for m # k
Xkm

B;, = - Bkm, for m # k

Bik = C Bkm
mok

The equations are solved alternatively using the most recent values of V and 0
available as shown in Fig. 4.8.(*)

, converged

NO
converged

Solve 4.10.9 and update ( 8 ) NO

\solve 4.10.10 and update (

J.
?.
]
converged

OUTPUT RESULTS

Fig. 4.8. Flow diagram of the fast-decoupled load flow

The matrices Bf and B" are real and are of order ( N - 1) and ( N - M )
respectively, where N is the number of busbars and M is the number of PV
busbars. B" is symmetric in value and so is B' if phase shifters are ignored; it is
found that the performance of the algorithm is not adversely affected. The
elements of the matrices are constant and need to be evaluated and triangulated
only once for a network.
Convergence is geometric, two to five iterations are required for practical
accuracies, and more reliable than the formal Newton's method. This is because
the elements of B' and B" are fixed approximations to the tangents of the defining
functions APIV and AQ/V, and are not susceptible to any 'humps' in the defining
functions.
If AP/V and AQ/V are computed efficiently, then the speed for iterations of the
fast decoupled method is about five times that of the formal Newton-Raphson or
about two-thirds that of the Gauss-Seidel method. Storage requirements are
about 60 percent of the formal Newton, but slightly more than the decoupled
Newton method.
Changes in system configurations are easily effected, and while adjusted
solutions take many more iterations these are short in time and the overall
solution time is still low.
The fast decoupled Newton load flow can be used in optimization studies for
a network and is particularly useful for accurate information of both real and
reactive power for multiple load-flow studies, as in contingency evaluation for
system security assessment.

4.11. CONVERGENCE CRITERIA AND TESTS(' 3,

The problem arises in the load-flow solution of deciding when the process has
converged with sufficient accuracy. In the general field of numerical analysis, the
accuracy of solution of any set of equations F(X) = 0 is tested by computing the
'residual' vector F(XP).The elements of this vector should all be suitably small for
adequate accuracy, but how small is to a large extent a matter of experience of the
requirements of the particular problem.
The normal criterion for convergence in load flow is that the busbar power-
mismatches should be small, i.e. AQi and/or APi, depending upon the type of
busbar i, and can take different forms, e.g.
IAPi(5 c, , for all PQ and PV busbars (4.11.1)

1 AQi1 5 c,, for all PQ busbars

where c, and c, are small empirical constants, and c, = c, usually. The value of c
used in practice varies from system to system and from problem to problem. In a
large system, c = 1 MW/MVAR typically gives reasonable accuracy for most
purposes. Higher accuracy, say c = 0.1 MW/MVAR may be needed for special
studies, such as load flows preceding transient stability calculations. In smaller
systems, or systems at light load, the value of c may be reduced. For approximate
load flows, c may be increased, but with some danger of obtaining a meaningless
solution if it becomes too large. Faced with this uncertainty, there is thus a
tendency to use smaller values of c than are strictly necessary. The criterion
(equation (4.11.1))is probably the most common in use. A popular variant on it is

and other similar expressions are also being used.

In the Newton-Raphson algorithms the calculation and testing of the
mismatches at each iteration, are part of the algorithm.
The set of equations defining the load-flow problem has multiple solutions,
only one of which corresponds to the physical mode of operation of the system. It
is extremely rare for there to be more than one solution in the neighbourhood of
+
the initial estimates for the busbar voltages ((1 jO) p.u., in the absence of anything
better), and apart from the possibility of data errors, a sensible-looking
mathematically-converged solution is normally accepted as being the correct
one. However, infrequent cases of very ill-conditioned networks and systems
operating close to their stability limits arise where two or more mathematically-
converged solutions of feasible appearance can be obtained by different choices of
starting voltages, or by different load-flow algorithms.
A load flow problem whose data corresponds to a physically unstable system
operating condition (often due to data errors, or in the investigation of unusual
operating modes, or in system planning studies) usually diverges. However, the
more powerful solution methods, and in particular Newton-Raphson, will
sometimes produce a converged solution, and it is not always easy in such cases
to recognize that the solution is a physically-unstable operating condition.
Certain simple checks, e.g. on the transmission angle and the voltage drop
across each line, can be included in the program to automatically monitor
the solution.
Finally, a practical load-flow program should include some automatic test to
discontinue the solution if it is diverging, to avoid unnecessary waste of
computation, and to avoid overflow in the computer. A suitable test is to check
at each iteration whether any voltage magnitude is outside the arbitrary range
0.5-1.5 p.u., since it is highly unlikely in any practical power system that a
meaningful voltage solution lies outside this range.

4.12. NUMERICAL EXAMPLE

A reduced version of the New Zealand system described in Appendix I1 is used
here to test the algorithm developed in this chapter.
The test network, illustrated in Fig. 4.9 involves the main generating and
loading points of the South Island, with the h.v.d.c. convertor represented as a
load, i.e. by specified P and Q.
The following computer print out illustrates the numerical input and output
information for the specified conditions.
 * U t li F
. tn
.................
CC U
00000000000000000

00000000000000000
OOOOOGOOOOOOOOOOO
C h UC

¶
I-
I
0
w h: r 0 w
(roo' 0 (D (rcn
00000000C000Ul000r I C
e................ <
00000000000000000 b tn
00000000000000030

C W N 0' G
m cno 9
00000000000000000 b

.................
00000000000000000

ccccccccCc~cc
ccccc Bq
00000000000000000
-I
zu
CO

x
z z z
C C
x
c m m
C
X
~ m m
I) a
x
.................
OGOOOOOOOOOOOGOOO

00000000000000000 X X
00000000000000000 C
xD-
CZ
Pm

...*..
CCCCLC
4

rl
OCOOOO 0
n
 0
a
'I-

1 Q
a 00000000000000000000 4
Z NNNNNNNNNNNNNNNNNNNN 3
WNNYNNWNNNNNNNNNNNNW n
>>ZZ>>><Jc<Z>'-'->~~dT4 Q.
zza~ZZzOOOW~>>~3L9u3 0
uus z u m a a a m J q a J t + I - r n c J 0
3
1 0
I : .
0
 eus DATA
GENERATION LOAO SHLNT
US N A M ~ U~JLTS WGLE MY HVAR rn MVAR MVPR

1 INV22O U.936 -1d.257 0.00 0.00 200.0U 51.00 0.00

10 AVIO11 1.045-27.414 200.00 115.40 0.0~ U.00 0.00

11 OHAU 1.060 -2>-433 350.00 113.38 0.0U 0.00 0.00

la LIV~~U u.966 -2q.266 0.00 0.00 150.0~ 60.00 0.00

THE MAXIMUM VIsMATcH I > 0.0041 ON BUS 3

THE SLACK BUS MNERATIII,. IS 7zd.709 a42.372
 L I N E ANU TRANSFORCER DATA

BUS hAMt VW WVPR

.....................................
3 MAN220
3
5
MAN220
TIC220
I
49'339
* r:g*%:$
39:647
2 Xb:%
MISCATCh
49'339
51:092
-0.014
3 9 47
-49:997
-0.004

3 MAN220
3 MAN220
1 INV220
I 60.722
I-49.244
60.722
-49.933
-49.933
-42.559
1 INV220 -49 244 -42.659
MISCATCh -0:067 -0.016

i! ROX220
MISVATCh

.....................................
7 BEN220 0.006 223.602
MISVATCh -0.006 0.000
.........................................................................
12 LIV220 2 1 370 92.023
7 BEhZ20 88'957 0.732
7 BEN220 88:957 0.732
10 A V I O l l -199.261 -93.686
MISV ATCh -0.023 -0.000
.........................................................................
4 AVI220 199.993 115.661
MISV A T C r 0.007 0.000

17 ~ ~ 1 2 2 0 350.004 113.378
MISCATCh -0.004 0.000

--I---------------------------------

13 W228
15 T E K O l l
ii3:118 -i8:0#
-149.320 12. 90
MISVATCh -0.007 0.001
 --i
ISL 220

OHA TWI 2 2 0

w AVI 011 AV1220

I
LIV 2 2 0

ROX 011 ROX 2 2 0

L
INV 2 2 0

MAN 220

TIW 2 2 0

Fig. 4.9. Reduced primary a.c. system for the South Island of
New Zeaiand

4.13. REFERENCES
1. B. Stott, 1974. 'Review of load-flow calculation methods', Proc. I EEE, 62 (7),9 16-929.
2. J. B. Ward and H. W. Hale, 1956. 'Digital computer solution of power-flow problems',
Trans. A1EE., PAS-75, 398-404.
3. H. E. Brown, G. K. Carter, H. H. Happ and C. E. Person, 1963. 'Power-flow solution
by impedance matrix iterative method', IEEE Trans. PAS-82, 1-10.
4. J. E. Van Ness and J. H. Griflin, 1961. 'Elimination methods for load-flow studies',
Pans. A1EE., PAS-80, 299-304.
5. W. F. Tinney, C. E. Hart, 'Power flow solution by Newton's method', IEEE Trans.,
PAS86 (1 l), 1449-1460.
6. E. C. Ogbuobiri, W. F. Tinney and J. W. Walker, 1970. 'Sparsity-directed decom-
position for Gaussian elimination on matrices', IEEE, Trans., PAS-89 (I), 141-150.
7. B. Stott and E. Hobson, 1971. 'Solution of large power-system networks by ordered
elimination: a comparison of ordering schemes', Proc. IEE, 118 (I), 125-134.
 8. W. F. Tinney and J. W. Walker, 1967. 'Direct solutions of sparse network equations
by optimally ordered triangular factorization', Proc. IE EE, 55 ( 1I), 1801-1 809.
9. S. T. Despotovic, 1974.'A new decoupled load-flow method', IEEE Trans., PAS93 (3),
884-89 1.
10. B. Stott, 1972. 'Decoupled Newton load flow', IEEE Trans., PAS-91, 1955-1959.
11. B. Stott and 0.Alsac, 1974. 'Fast decoupled load flow', IEEE Trans., PAS93 (3), 859-
869.
12. J. Medanic and B. Avramovic, 1975.'Solution of load-flow problems in power systems
by &-couplingmethod', Proc IEE, 122 (8), 801-805.
13. B. Stott, 1972. 'Power-system load flow' (M.Sc. lecture notes). University of
Manchester Institute of Science and Technology.
 Three-Phase Load Flow
5.1. INTRODUCTION
For most purposes in the steady-state analysis of power systems, the system
unbalance can be ignored and the single-phase analysis described in Chapter 4 is
adequate. However, in practice it is uneconomical to balance the load completely
or to achieve perfectly balanced transmission system impedances, as a result of
untransposed high-voltage lines and lines sharing the same right of way for
considerable lengths.
Among the effects of power-system unbalance are: negative sequence currents
causing machine rotor overheating, zero sequence currents causing relay
maloperations and increased losses due to parallel untransposed lines.
The use of long-distance transmission motivated the development of analytical
techniques for the assessment of power-system unbalance. Early techniques(').(2)
were restricted to the case of isolated unbalanced lines operating from known
terminal conditions. However, a realistic assessment of the unbalanced operation
of an interconnected system, including the influence of any significant load
unbalance, requires the use of three-phase load-flow a l g o r i t h m ~ . ( ~ ) . (The
~)?(~)
object of the three-phase load flow is to find the state of the three-phase power
system under the specified conditions of load, generation and system
configuration.
The basic three-phase models of system plant and the rules for their
combination into overall network admittance matrices, discussed in Chapter 2,
are used as the framework for the three-phase load flow described in this chapter.
The storage and computational requirements of a three-phase load-flow
program are much greater than those of the corresponding single-phase case. The
need for efficient algorithms is therefore significant even though, in contrast to
single-phase analysis, the three-phase load flow is likely to remain a planning,
rather than an operational exercise.
The basic characteristics of the Fast Decoupled Newton-Raphson algorithms
described in Chapter 4, have recently been shown(@to apply equally to the three-
phase load-flow problem. Consequently, this algorithm is now used as a basis
for the development of an efficient three-phase load-flow program. When the
program is used for post-operational studies of important unbalanced situations
. ... , I
1

w:3j3 '-*+ +.h,.-f:. 90

. ' l 9r '..
.'- ., . . .
on the power system, additional practical features such as automatic transformer
tapping and generator VAR limiting are necessary.

5.2. NOTATION
A clear and unambiguous identification of the three-phase vector and matrix
elements requires a suitable symbolic notation using superscripts and subscripts.
The a.c. system is considered to have a total of n busbars where:
+
n = nb ng
n b is the number of actual system busbars
ng is the number of synchronous machines
Subscripts i, j, etc. refer to system busbars as shown in the following examples:
i = 1, nb identifies all actual system busbars, i.e. all load busbars plus all
generator terminal busbars.
i = n b + 1, nb + ng - 1 identifies all generator internal busbars with the
exception of the slack machlne.
i = n b + ng identifies the internal busbar at the slack machine.
The following subscripts are also used for clarity:
reg-refers to a voltage regulator
int-refers to an internal busbar at a generator
gen-refers to a generator
Superscripts p, m identify the three phases at a particular busbar.

5.3. FORMULATION O F THE THREE-PHASE LOAD-FLOW PROBLEM

Synchronous machine modelling

Synchronous machines are designed for maximum symmetry of the phase
windings and are therefore adequately modelled by their sequence impedances.
Such impedances contain all the information that is required to analyse the
steady-state unbalanced behaviour of the synchronous machine.
The representation of the generator in phase components may be derived from
the sequence impedance matrix (Z,), as follows:

where
and 'a' is the complex operator ej2"I3.The phase component impedance matrix is
thus

The phase component model of the generator is illustrated in Fig. 5.l(a) The
machine excitation acts symmetrically on the three phases and the voltages at the
internal or excitation busbar form a balanced three-phase set, i.e.

and

For three-phase load flow the voltage regulator must be accurately modelled as
it influences the machine operation under unbalanced conditions. The voltage
regulator monitors the terminal voltages of the machine and controls the
excitation voltage according to some predetermined function of the terminal
voltages. Often the positive sequence is extracted from the three-phase voltage
measurement using a sequence filter.
Before proceeding further it is instructive to consider the generator modelling
from a symmetrical component frame of reference. The sequence network model
of the generator is illustrated in Fig. 5.l(b). As the machine excitation acts
symmetrically on the three-phases, positive sequence voltages only are present at
the internal busbar.
The influence of the generator upon the unbalanced system is known if the
voltages at the terminal busbar are known. In terms of sequence voltages, the
positive sequence voltage may be obtained from the excitation and the positive
sequence voltage drop caused by the flow of positive sequence currents through
the positive sequence reactance. The negative and zero sequence voltages are
derived from the flow of their respective currents through their respective
impedances. It is important to note that the negative and zero sequence voltages
are not influenced by the excitation or positive sequence impedance.
There are infinite combinations of machine excitation and machine positive
sequence reactance which will satisfy the conditions at the machine terminals and
give the correct positive sequence voltage. Whenever the machine excitation must
be known (as in fault studies) the actual positive sequence impedance must be
used. For load flow however, the excitation is not of any particular interest and
the positive sequence impedance may be arbitrarily assigned to any value.@)The
positive sequence impedance is usually set to zero for these studies.
 I V term,
I f v e sequence

-ve sequence

zero sequence

Fig. 5.1. Synchronous machine models. (a) Phase component repre-

sentation. @) Symmetrical component representation

Thus the practice with regard to three-phase load flow in phase coordinates, is
to set the positive sequence reactance to a small value in order to reduce the
excitation voltage to the same order as the usual system voltages with a
corresponding reduction in the angle between the internal busbar and the
terminal busbar. Both these features are important when a fast decoupled
algorithm is used.
Therefore, in forming the phase component generator. model using equation
(5.3.4), an arbitrary value may be used for Z , but the actual values are used for 2,
and 2,. There is no loss of relevant information as the influence of the generator
upon the unbalanced system is accurately modelled.
The nodal admittance matrix, relating the injected currents at the generator
busbars to their nodal voltages, is given by the inverse of the series impedance
matrix derived from equation (5.3.4).

Specified variables
The following variables form a minimum and sufficient set to define the three-
phase system under steady-state operation:
- The slack generator internal busbar voltage magnitude Vin,j where j =
+
nb ng. (The angle Ointj is taken as a reference.)
- The internal busbar voltage magnitude Vintj and angles Ointj at all other

+
generators, i.e. j = n b 1, n b + ng - 1 .
- The three voltage magnitudes (Vf') and angles (Of)at every generator terminal

busbar and every load busbar in the system, i.e. i = 1, n b and p = 1,3.
Only two variables are associated with each generator internal busbar as the
three-phase voltages are balanced and there is no need for retaining the
redundant voltages and angles as variables. However, these variables are retained
to facilitate the calculation of the real and reactive power mismatches. The
equations necessary to solve for the above set of variables are derived from the
specified operating conditions, i.e.
- The individual phase real and reactive power loading at every system busbar.
- The voltage regulator specification for every synchronous machine.
- The total real power generation of each synchronous machine, with the

exception of the slack machine.

The usual load-flow specification of a slack machine, i.e. fixed voltage in phase
and magnitude, is applicable to the three-phase load flow.

Derivation of equations
The three-phase system behaviour is described by the equation

where the system admittance matrix [Y], as developed in Chapter 2, represents

each phase independently and models all inductive and capacitive mutual
couplings between phases and between circuits. The mathematical statement of
the specified conditions is derived in terms of the system admittance matrix
[ Y l = [GI +jP1
as follows:
(i) For each of the three phases (p) at every load and generator terminal
busbar (i),
A Pf = ( p f y - Pf
 = (PfYp- Vf
k=l m=l
V; [G$" cos O
r + BB sin O f l
and

= (Qf)""- Vf
n

k= 1 m= 1
x
3
V r [Gim sin Of7- B f j cos 0im] (5.3.9)

(ii) For every generator j,

where k is the bus number of the jth generator's terminal busbar.

(iii) For every generator j, with the exception of the slack machine, i.e. j + nb
+ ng.

= - z xz
p=l
3
Vint
n 3

k=l m=l
V r [Gjql"cos 04r + Bjql"sin 0 4 3
(5.3.11)
where, although the summation for k is over all system busbars, the
mutual terms Gj, and Bjkare nonzero only when k is the terminal busbar of
the jth generator.
It should be noted that the real power specified for the generator is the total real
power at the internal or excitation busbar whereas in actual practice the specified
quantity is the power leaving the terminal busbar. This in effect means that the
generator's real power loss is ignored.
The generator losses have no significant influence on the system operation and
may be calculated from the sequence impedances at the end of the load-flow
solution, when all generator sequence currents have been found. Any other
method would require the real power mismatch to be written at busbars remote
from the variable in question, that is, the angle at the internal busbar. In addition,
inspection of equations (5.3.8) and (5.3.1 1) will show that the equations are
identical except for the summation over the three phases at the generator internal
busbar.
That is, the sum of the powers leaving the generator may be calculated in
exactly the same way and by the same subroutines as the power mismatches at
other system busbars. This is possible because the generator internal busbar is
not connected to any other element in the system. Inspection of the Jacobian
submatrices derived later will show that this feature is retained throughout the
study. In terms of programming the generators present no additional complexity.
Equations (5.3.8) to (5.3.11) form the mathematical formulation of the three-
phase load flow as a set of independent algebraic equations in terms of the system
variables.
The solution to the load-flow problem is the set of 'variables which, upon
substitution, make the left-hand side mismatches in equations (5.3.8) to (5.3.1 1)
equal to zero.
5.4. FAST DECOUPLED THREE-PHASE ALGORITHM
The standard Newton-Raphson algorithm may be used to solve equations (5.3.8)
to (5.3.11). This involves an iterative solution of the matrix equation

for the right-hand side vector of variable updates. The right-hand side matrix in
equation (5.4.1) is the Jacobian matrix of first order partial derivatives.
Following decoupled single-phase load-flow practice, the effects of A 8 on
reactive power flows and A V on real power flows are ignored. Equation (5.4.1)
may therefore be simplified by assigning

and
[C]=[GI = O
In addition, the voltage regulator specification is assumed to be in terms of the
terminal voltage magnitudes only and therefore,
[Dl=[HI = 0
Equation (5.4.1) may then be written in decoupled form as:

for i, k = i, nb and j, 1 = 1, ng - 1 (i.e. excluding the slack generator)

for i, k = 1, nb and j, 1 = l,ng (i.e. including the slack generator).

To enable further development of the algorithm it is necessary to consider the
Jacobian submatrices in more detail.
In deriving these Jacobians from equations (5.3.8) to (5.3.11) it must be
remembered that,

when 1 refers to a generator internal busbar.

The coeficients of matrix equation (5.4.2) are:
or
ALm = V f 'V r [Gf'p sin Of'? - Bf'p cos Of?]
except for

where [Fill = 0 for all j # 1because the jth generator has no connection with the
lth generator's internal busbar, and

p= 1
3 3
+ m2= 2 (Vin,J2[Gf;'"sin O
1 p=l
r - BEmcos Of;"]
m f P

The coefficients of matrix equation (5.4.3) are

where
K$" = V r Vf [Gfp sin Of; - Bfp cos O$"]
except
K g " = - B;:'(v;)2 + Q;:

let [L;] = V;[L;]' where k is the terminal busbar of the jth generator and
L$ = 0 otherwise.

=0 for all j, 1 as the voltage regulator

specification does not explicitly
include the variables Vini.
Although the above expressions appear complex, their meaning and derivation
are similar to those of'the usual single phase Jacobian elements.

Jacobian approximations
Approximations similar to those applied to the single-phase load flow are
applicable to the Jacobian elements as follows:
(i) at all nodes (i.e. all phases of all busbars)

(ii) between connected nodes of the same phase,

cos 8;'" z 1 i.e. O,"," is small

(iii) Moreover the phase-angle unbalance at any busbar will be small and
hence an additional approximation applies to the three-phase system, i.e.

z f 120" for p # m
(iv) Finally, as a result of (ii) and (iii) the angle between different phases of
connected busbars will be approximately 120", i.e.

or
cos @" z - 0.5
and
sin Of? z 0.866
These values are modified for the f 30" phase-shift inherent in the
star-delta connection of three-phase transformers.
The final approximation (iv), necessary if the Jacobians are to be kept constant,
is the least valid, as the cosine and sine values change rapidly with small angle
variations around 120 degrees. This accounts for the slower convergence of the
phase unbalance at busbars as compared with that of the voltage magnitudes and
angles.
It should be emphasised that these approximations apply to the Jacobian
elements only, i.e. they do not prejudice the accuracy of the solution nor do they
restrict the type of problem which may be attempted.
Applying approximations (i) to (iv) to the Jacobians and substituting into
equations (5.4.2) and (5.4.3) yields,
and

where

with

8;r = 120" for p # m

All terms in the matrix [MI are constant, being derived solely from the system
admittance matrix. Matrix [MI is the same as matrix [ - B] except for the off-
diagonal terms which connect nodes of different phases. These are modified by
allowing for the nominal 120" angle and also including the GIP,"sin 0: terms.
The similarity in structure of all Jacobian submatrices reduces the pro-
gramming complexity normally found in three-phase load flows. This uniformity
has been achieved primarily by the method used to implement the three-phase
generator constraints.
The above derivation closely parallels the single-phase fast decoupled
algorithm, but the added complexity of the notation obscures this feature. At the
present stage the Jacobian elements in equations (5.4.4) and (5.4.5) are identical
except for those terms which involve the additional features of the generator
modelling.
These functions are more linear in terms of the voltage magnitude [qthan are
the functions CAP] and [AQ]. In the Newton-Raphson and related constant
Jacobian methods the reliability and speed of c~nvergence'~) improve with the
linearity of the defining functiods. With this aim, equations (5.4.4) and (5.4.5) are
modified as follows :
- The left-hand side defining functions are redefined as [APrlVf],
[Apeenjl vin,jl and CAQPIVPI
- In equation (5.4.4), the remaining right-hand side V terms are set to 1 p.u.
- In equation (5.43, the remaining right-hand side V terms are cancelled by the

corresponding terms in the right-hand side vector.

 These modifications yield the following expressions:

Recalling that [LEI' = [aA Vreg,/a V,"], as Vreg is normally a simple linear
function of the terminal voltages, [L'] will be a constant matrix.
Therefore, the Jacobian matrices [B'] and [B"] in equations (5.4.6) and (5.4.7)
have been approximated to constants.
Zero diagonal elements in equation (5.4.7) may result from the ordering of the
equations and variables. This feature causes no problems if these diagonals are
not used as pivots until the rest of the matrix has been factorized (by which time,
fill-in terms will have appeared on the diagonal). This causes a minor loss of
efficiency as it inhibits optimal ordering for the complete matrix. Although this
could be avoided by reordering the equations, the extra program complexity is
not justified.
Based on the reasoning of Stott and Alsac,") which proved successful in the
single-phase load flow the [B'] matrix in equation (5.4.6) is further modified by
omitting the representation of those elements that predominantly affect MVAR
flows.
The capacitance matrix and its physical significance is illustrated in Fig. 5.2, for
a single three-phase line. With n capacitively-coupled parallel lines the matrix will
be 3n x 3n.
In single-phase load flows the shunt capacitance is the positive sequence
capacitance which is determined from both the phase-to-phase and the phase-to-
earth capacitances of the line. It therefore appears that the entire shunt
capacitance matrix predominantly affects MVAR flows only. Thus, following
single-phase fast decoupling practice the representation of the entire shunt
capacitance matrix is omitted in the formulation of [B']. This increases
dramatically the rate of real power convergence.
With capacitively coupled three-phase lines the interline capacitance in-
fluences the positive sequence shunt capacitance. However, as the values of
interline capacitances are small in comparison with the self capacitance of the
phases, their inclhsion makes no noticeable difference. The effective tap of fi
introduced by the star-delta transformer connection is interpreted as a nominal
tap and is therefore included when forming the [B'] matrix.
 ( ii )

Fig. 5.2. Shunt capacitance matrices

==F=T
Calculate [AO/V],[AV,, ] Output

-
Fig. 5.3. Iteration sequence for three-phase a.c. load flow
 A further difficulty arises from the modelling of the star-gldelta transformer
connection. The equivalent circuit, illustrated in Section 2.6 shows that large
shunt admittances are effectively introduced into the system. When these are
excluded from [B'], as for a normal shunt element, divergence results. The entire
transformer model, must therefore be included in both [B'] and [B"].
With the modifications described above the two final algorithmic equations
may be concisely written, i.e.

The constant Jacobians [Bk] and [Bi] correspond to fixed approximated

tangent slopes to the multidimensional surfaces defined by the left hand-side
defining functions.
Equations (5.4.8) and (5.4.9) are then solved according to the iteration sequence
illustrated in Fig. 5.3.

Generator models and the fast decoupled algorithm

The derivation of the fast decoupled algorithm involves the use of several
assumptions to enable the Jacobian matrices to be approximated to constant.
The same assumptions have been applied to the excitation busbars associated
with the generator model as are applied to the usual system busbars. The validity
of the assumptions regarding voltage magnitudes and the angles between
connected busbars depends upon the machine loading and positive sequence
reactance. As discussed in Section 5.3 this reactance may be set to any value
without altering the load-flow solution and a value may therefore be selected to
give the best algorithmic performance.
When the actual value of positive sequence reactance is used the angle across
the generator and the magnitude of the excitation voltage both become
comparatively large under full-load operation. Angles in excess of forty-five
degrees and excitation voltages greater than 2.0 p.u. are not uncommon. Despite
this considerable divergence from assumed conditions, convergence is surpris-
ingly good. Convergence difficulties may occur at the slack generator and then
only when it is modelled with a high synchronous reactance (1.5 p.u. on machine
rating) and with greater than 70 percent full load power.
All other system generators, where the real power is specified, converge reliably
but somewhat slowly under similar conditions.
This deterioration in convergence rate and the limitation on the slack
generator loading may be avoided by setting the generator positive sequence
reactance to an artificially low value (say 0.01 p.u. on machine rating), a
procedure which does not involve any loss of relevant system information.
5.5. STRUCTURE OF THE COMPUTER PROGRAM
The main components of the computer program are illustrated in Fig. 5.4. The
approximate number of FORTRAN statements for each block is indicated in
parenthesis. The main features of each block are described in the following
sections.

Data input
The input data routine implements the system modelling techniques described in
Chapter 2 and forms the system admittance model from the raw data for each
system component. Examples of the raw data are given in Section 5.7. with
reference to a particular test system.
The structure and content of the constant Jacobians B' and B" are based upon
the system admittance matrix and are thus formed simultaneously with this
matrix.
Both the system admittance matrix and the Jacobian matrices are stored and
processed using sparsity techniques which are structured in 3 x 3 matrix blocks
to take full advantage of the inherent block structure of the three-phase system
matrices.
Data Input
Form and store system
admittance matrix, [ B ' ]
and [ B " ]Jacobian
I matrices. (1645)
I
1

I Z Z i z e [ B ' ]and [ B * ]
I
I Initialize three-phase
system voltages (200) I
procedure (see Fig. 5.3)
(450)

voltages, line power flows

and total system losses
(215)
- --

Fig. 5.4. Program structure

Factorization of constant Jacobians
The heart of the load-flow program is the repeat solutions of equations (5.4.8) and
(5.4.9) as illustrated in Fig. 5.3. These equations are solved using sparsity
techniques and near optimal ordering as discussed in Chapter 4 (Section 4.7) or
like those embodied in Zollenkopfs bifa~torization.(~) The constant Jacobians
are factorized before the iteration sequence is initiated. The solution of each
equation within the iterative procedure is relatively fast, consisting only of the
forward and back substitution processes.

Starting values
Starting values are assigned as follows:
- The nonvoltage controlled busbars are assigned 1 p.u. on all phases.
- At generator terminal busbars all voltages are assigned values according to
the voltage regulator specifications.
- All system busbar angles are assigned 0, - 120°, + 120" for the three phases
respectively.
- The generator internal voltages and angles are calculated from the specified
real power and, in the absence of better estimates, by assuming zero reactive
power. For the slack machine the real power is estimated as the difference
between total load and total generation plus a small percentage(say 8 percent)
of the total load to allow for losses.
For cases where convergence is excessively slow or difficult it is recommended
to use the results of a single-phase load flow to establish the starting values. The
values will, under normal steady-state unbalance, provide excellent estimates for
all voltages and angles including generator internal conditions which are
calculated from the single-phase real and reactive power conditions.
Moreover, as a three-phase iteration is more costly than a single-phase
iteration, this practice can be generally recommended to provide more efficient
overall convergence and to enable the more obvious data errors to be detected at
an early stage.
For the purpose of investigating the load-flow performance, flat voltage and
angle values are used in the examples that follow.

Iterative solution
The iterative solution process (Fig. 5.3) yields the values of the system voltages
which satisfy the specified system conditions of load, generation and system
configuration.

Output results
The three-phase busbar voltages, the line power flows and the total system losses
are calculated and printed out. An example is given in Table 5.6. In addition the
sequence components of busbar voltages are also calculated as these provide a
more direct measure of the unbalance present in the system under study.

5.6. PERFORMANCE OF THE ALGORITHM

This section attempts to identify those features which influence the convergence
with particular reference to several small to medium-sized test systems.
The performance of the 'three-phase' algorithm is examined under both
balanced and unbalanced conditions, and comparisons are made with the
performance of the single-phase fast-decoupled algorithm.

Performance under balanced conditions

A symmetrical three-phase system, operating with balanced loading, is accurately
modelled by the positive sequence system and either a three-phase or a single-
phase load flow may be used to analyse the system. Under these conditions it is
possible to compare the performance of the three-phase and single-phase fast-
decoupled algorithms.
The three-phase system transmission lines are represented by balanced full 3
x 3 matrices. Transformers are modelled with balanced parameters on all phases
and generators are modelled by their phase parameter matrices as derived from
their sequence impedances.
Typical numbers of iterations to convergence for both the single-phase and
three-phase algorithms, given in Table 5.1, indicate that the algorithms behave
identically. Features such as the transformer connection and the negative and
zero sequence generator impedances have no effect on the convergence rate of the
three-phase system under balanced conditions. This is not unexpected as, under
balanced conditions, only the positive sequence network has any power flow and
there is no coupling between sequence networks. The negative and zero sequence
information inherent in the three-phase system model of the balanced systems,
has no influence on system operation and this is reflected into the performance of
the algorithm.

Table 5.1. Convergence results

Number Single- Balanced three

of phase phase load flow Typical three-
Case busbars load flow II M phase unbalance

4,3 4,3 4,3 66

3-3 3,3 3-3 8,8
3,3 3,3 3,3 6,5
3,3 3,3 3,3 8,7
393 3,3 393 696
Convergence tolerance is 0.1 MWIMVAR. The numerical results, (i,~], should be interpreted as
follows:
i-refers to the number of real power-angle update iterations.
j-refers to the number of reactive power-voltage update iterations.
MW f ( a 1 Real power mismatches(p.u.)
( p.u.1
MVAr
( p.u.1 t ( b ) Reoctive power mismatches

ltemtion lterition

Fig. 5.5. Power convergence patterns for three-phase and single-phase load flow

VoltaJe
(plr.1 A
1.04 Three-phase busbar
voltages
3

2
2

1 1'
I I t
2 3 4 5 Iteration
(i 1

Voltage
( pu.)

Single-phase load flow

Fig. 5.6. Voltage convergence patterns for three-phase and

single-phase load flows. (i) Three-phase voltages. (ii) Single-
phase and three-phase positive sequence voltages
Performance with unbalanced systems
The number of iterations to convergence for the same test systems, under realistic
steady state unbalanced operation, are also given in Table 5.1. The convergence
rate deteriorates as compared with the balanced case, requiring on average twice
as many iterations.
The graphs of Fig. 5.5 show that initial convergence of the three-phase
mismatches is very close to that of the single-phase load flow. However, as the
solution is approached the three-phase convergence becomes slower. It appears
that, although the voltage and angle unbalance are introduced from the first
iteration, they have only a secondary effect on the convergence until the positive
sequence power flows are approaching convergence.

Generator

Synchronous
condenser

Fig. 5.7. Test system single-line diagram

 This feature is further illustrated in Fig. 5 4 9 where the convergence pattern of
the three-phase voltages is shown. The convergence pattern of the positive
sequence component of the unbalanced voltages is shown in Fig. 5.6(ii) together
with the convergence pattern of the voltage at the same busbar for the
corresponding single-phase load flow. The latter figure illustrates that the
positive sequence voItage of the three-phase unbalanced load flow has an almost
identical convergence pattern to the corresponding single-phase fast-decoupled
load flow.

Genemtor
excitation voltopes

-r
&
MAN GN

MAN M4

Excitation
voltages

Fig. 5.8. Test system 3 x 3 compound coil representation

 The final convergence of the system unbalance is somewhat slow but is reliable.
The following features are peculiar to a three-phase load flow and their
influence on convergence is of interest:
- Asymmetry of the system parameters.
- Unbalance of the system loading.
- Influence of the transformer connection.
- Mutual coupling between parallel transmission lines.
These features have been examined with reference to a small six-bus test system.

Secondory ( S 1

Primary ( P
MAN 220

0
Fig. 5.9. Test system 3 x 3 matrix representation
5.7. TEST SYSTEM AND RESULTS
A single-line diagram of the test system under consideration is illustrated in
Fig. 5.7. Some features of interest are:
- An example of a line sectionalization is included. One section contains four
mutually coupled three-phase power lines. The other section contains two
sets of two mutually coupled three-phase lines.
- All parallel lines are represented in their unbalanced mutually coupled state.
- Both transformers are star-delta connected with the star neutrals solidly
earthed. Tap ratios are present on both primary and secondary sides.
The system is redrawn in Fig. 5.8 using 3 x 3 compound coil notation and
substituting for the generator and line models. Following this, Fig. 5.9 illustrates
the system graphically in terms of 3 x 3, 6 x 6 and 12 x 12 matrix blocks,

Fig. 5.10. Test system exploded into eight systems

representing the various system elements. The matrix quantities illustrated in
Fig. 5.9 are given by, or derived from, the input data to the load-flow program.
For the purposes of input data organization and the formation of the system
admittance matrix, the system is divided into eight natural subsystems. These are
illustrated in the exploded system diagram of Fig. 5.10.
Once the matrices defined in Figs. 5.9 and 5.10 are known, the admittance
matrix for each subsystem can be formed following the procedures outlined in
Chapter 2. The subsystems are then combined to form the overall system
admittance matrix.
The input data, which enables all the matrices in Fig. 5.10 to be formed, is listed
for each subsystem in the following sections. The data is all in p.u. to a base of 33.3
MVA.

Generator data-subsystems 1 and 2

Subsystems 1 and 2 represent two synchronous generators. The input data to the
computer program consists of the three-sequence impedances, the voltage
regulator specification and the total real power generation at all generators
except one which is the slack machine.
Table of generator data

Generator Zero Impedance Pos. Impedance Neg. Impedance P Voltage

No. Name RO XO R1 X1 R2 X2 p.u. regulator
'phase

1MAN014 0.0 0.080 0.0 0.010 0.0 0.021 15.000 1.045

2 ROXOll 0.0 0.150 0.0 0.010 0.0 0.091 SLACK 1.050

The effect of subsystem 3 (the synchronous condenser) is not included in the numerical example

Transformer data-subsystems 4 and 5

The input data for a transformer subsystem consists of:
(a) Leakage impedance in p.u. (r j x ) . +
(b) Transformer type (specified in case descriptions).
(c) Primary and secondary tap ratios.
The data for the two transformers is summarized in the following table.

Busbar names
Leakage Tap ratio
primary secondary reactance prim.

MAN220 MAN014 0.0006 +j0.0164 0.045

ROX220 ROXOll 0.0020 +j0.038 0.022
- - -

The subsystem matrices are formed as described in Section 2.6.

Line data-subsystems 6, 7 and 8

The series impedance and shunt admittance matrices must be read into the
computer program.
Subsystem 6
This subsystem consists of a single-balanced line between the two terminal
busbars. The phases are taken as uncoupled and the matrices are given below.
Terminal Busbars INV 220-ROX 220

[Z,] =b I 0.002 +fl.O15 I 0.006 +jO.050 I 0.002 +jO.017 I

Both these matrices are in p.u. for the total line length.
Subsystem 7
This subsystem consists of a pair of parallel, mutually coupled three-phase lines.
These lines are represented in their-natural coupled unbalanced state.
Terminal busbars
Line 1 INV220 - TIW220
Line 2 INV220 - TIW220
The series impedance matrix for the total length (2,) is:
Line 1 Line 2
b b

0.0012
Line 1 b
+jO.008

Line 2 b
+jO.0061
The shunt admittance matrix for the total line length is:
Line 1 Line 2
a b c a b c

Line 1 b

Line 2 b

The lower diagonal half only is shown as all line matrices are symmetrical.
Subsystem 8
Sectionalized mutually coupled lines: Section 1 consists of four mutually coupled
three-phase lines and has 12 x 12 characteristic matrices, [Z,,] and [ Y , , ] , as
indicated in the system diagrams. These are given in Figs. 5.1 1 and 5.12 in per unit
length of line and Section 1 is taken as having a length of 0.75 units.
Section 2 consists of two sets of two mutually coupled three-phase lines. To
ensure consistent dimensionality with Section 1, the second section is considered
as being composed of four mutually coupled three-phase lines, the elements
representing the coupling between the two separate double-circuit lines being set
to zero. The characteristic matrices for Section 2 become,

Section 2 4

1 2 3 4
c Y*I
12 x 12
Section 2

where 101 is a matrix of zeros. The submatrices are labelled as those in Fig. 5.8.
These 12 x 12 matrices are given in Fig. 5.13 and Fig. 5.14. in per unit length of
line. Section 2 is taken as having a length of 0.25 units.
Once the overall admittance matrix for the combined sections has been found
it must be stored in full. This is to enable calculation of the power flows in the four
individual lines. The matrix is modified, as described in Section 2.5. for the
terminal connections. This modified matrix is the subsystem admittance matrix
to be combined into the overall system admittance matrix.
 Line 1 Line 4

Line I b

Line 2 b

Line 3 b

Line 4 b

Fig. 5.12. Shunt admittance matrix [ Y , , ] for Section 1

118
Test cases and typical results
The following cases have been examined :
Balanced system with balanced loading and no mutual coupling between
parallel three-phase lines. Generator transformers are star-glstar-g.
As for case (i) but with balanced mutual coupling introduced for all
parallel three-phase lines as indicated in Fig. 5.7.
As for case (ii) but with balanced loading.
As for case (ii) but with system unbalance introduced by line capacitance
unbalance only.
As for case (ii) but with system unbalance introduced by line series
impedance unbalance only.
Combined system capacitance and series impedance unbalance with
balanced loading. Generator transformers star-glstar-g.
As for case (vi) but with unbalanced loading.
As for case (vii) but with deltalstar-g .for the generator transformers.
As for case (viii) but with large unbalanced real power loading at INV220.
As for case (viii) but with large unbalanced reactive power loading at
INV220.
The number of iterations to convergence, given in Table 5.2, clearly indicates
that system unbalance causes a deterioration in convergence. Such deterioration
is largely independent of the cause of the unbalance, but it is very dependent on
the severity or degree of the unbalance.
In all these cases the degree of system unbalance is significant as may be
assessed from the sequence components of the busbar voltages, which are given in
Table 5.3 for cases (vii), (viii), and (x). The latter case is only included to
demonstrate the convergence properties of the algorithm.
It is noteworthy that the initial convergence of the algorithm is fast even in cases
of extreme steady-state unbalance. The reliability of the algorithm is not

Table 5.2. Number of iterations to convergence for six

bus test system

Convergence tolerance (MWIMVAR)

Case 10.0 1O
. 0.1

i
11
...
111
iv
v
vi
vii
viii
ix
X
 Table 5.3. Sequence components of busbar voltages

+ ve sequence - ve sequence Zero sequence

Busbar VI 01 v2 02 vo 00
INV220 1.020 - 0.16 0.028 2.42 0.021 - 0.85
ROX220 1.037 - 0.13 0.028 2.37 0.025 - 1.13
MAN220 1.058 7 0.09 0.01 5 1.84 0.014 - 0.77
MAN0 14 1.039 - 0.01 0.008 1.85 0.012 - 0.76
TIW220 1.015 - 0.17 0.028 2.40 0.021 - 0.74
ROXOl 1 1.055 - 0.03 0.019 2.39 0.019 - 1.12
MAN.GN 1.056 0.03 0.0 - 0.0 -
ROX.GN 1.066 0.0 0.0 - 0.0 -

Case (vii)

+ ve sequence - ve sequence Zero sequence

Busbar VI 01 v2 02 vo 00
INV220 0.023 - 3.12 0.004 0.23
ROX220 0.023 3.04 0.005 - 0.80
MAN220 0.015 2.39 0.001 0.20
MAN014 0.006 2.93 0.0 -
TIW220 0.023 3.11 0.005 0.69
ROXO 1 1 0.016 - 2.70 0.0 -
MAN.GN 0.0 A
0.0 -
ROX.GN 0.0 - 0.0 -

Case (viii)

+ ve sequence - ve sequence Zero sequence

Busbar VI 01 v2 02 vo 00

prejudiced by significant unbalance although convergence to small tolerances

becomes slow.
The influence of the three-phase transformer connection may be seen in the
sequence voltages of cases (vii) and (viii). The star-gldelta connection provides no
through path for zero-sequence currents and the zero-sequence machine current
is zero. This is reflected in the zero-sequence voltages at the machine terminal
voltages.
The sequence voltages also illustrate the position of angle reference at the slack
generator internal busbar. In addition, it may be seen that at all generator
internal busbars the negative and zero-sequence voltages are zero reflecting the
balanced and symmetrical nature of the machine excitations.
As an example of the numerical results, the busbar loadings for case (viii) are
given in Table 5.4 and the resulting busbar voltages and line power flows are
presented in Tables 5.5 and 5.6.
Besides the significant unbalance other features to be noticed are:
d

- The approximate 30" phase shift due to the star-delta connected

transformers.
- Balanced voltages at the generator-internal busbars.
- Balanced angles at the generator-internal busbars.

- An apparent gain in active power flow in any one phase. This power flows
through the mutual coupling terms between phases. The overall active power
shows a net loss as expected for a realistic system.

Table 5.4. Table of busbar data

Busbar Phase A Phase B Phase C

NO. Name P-Load Q-Load P-Load Q-Load P-Load Q-Load

1 INV220 50.000 15.000 45.000 14.000 48.300 16.600

2 ROX220 48.000 20.000 47.000 12.000 51.300 28.300
3 MAN220 0.0 0.0 0.0 0.0 0.0 0.0
4 MAN014 0.0 0.0 0.0 0.0 0.0 0.0
5 TIW220 150.000 80.000 157.000 78.000 173.000 72.000
6 ROXOll 0.0 0.0 0.0 0.0 0.0 0.0

Table 5.5. Busbar results

Busbar Phase A Phase B Phase C Generation

No. Name Volt Ang Volt Ang Volt Ang Total

INV220
ROX220
MAN220
MAN014
TIW220
ROXO 11
MAN.GN
ROX. GN
 Table 5.6 Computed print-out power flows

Sending end Receiving end

Sending end Busbar Receiving end Busbar
MW MVAR MW MVAR

No. Name No. Name

4 MAN014 7 M AN.GN - 163.583 - 62.676 164.077 71.179
- 160.184 - 47.925 159.968 55.047
- 176.232 - 50.050 175.955 59.577
6 ROXO 11 8 ROX.GN - 95.41 6 - 37.762 96.329 41.642
- 87.270 - 34.303 87.620 35.449
- 98.590 - 27.893 97.327 31.011
 C
t4
t4
Table 5.6. (contd.)
- --

Sending end Receiving end

Sending end Busbar Receiving end Busbar
MW MVAR MW MVAR

3 MAN220 4 MAN014 - 157.242 -42.113 163.587 62.660

- 166.786 - 31.943 160.186 47.958
- 174.468 - 45.462 176.229 50.033
2 ROX220 6 ROXOl 1 - 92.984 - 26.544 95.416 37.757
- 87.935 - 17.105 87.271 34.312
- 98.772 - 25.566 98.588 27.888

Total Generation 781.27 MW 293.91 MVAR

Total Load 768.60 MW 335.90 MVAR
System Losses 11.67 MW - 41.98 MVAR

Mismatch 0.001 3 MW - 0.0096 MVAR

5.8. REFERENCES
1. A. Holley, C. Coleman and R. B. Shipley, 1964. 'Untransposed e.h.v. line com-
putations', Trans. I EEE, PAS-83, 291.
2. M. H. Hesse, 1966. 'Circulating currents in parallel untransposed multicircuit lines: I-
Numberical evaluations', Trans. IEEE, PAS-85, 802. '11-Methods of eliminating
current unbalance', Trans. I EEE, PAS95, 812.
3. A. H. El-Abiad and D. C. Tarisi, 1967. 'Load-flow solution of untransposed EHV
networks', PICA, Pittsburgh, Pa, 377-384.
4. R. G. Wasley and M. A. Slash, 1974. 'Newton-Raphson Algorithm for three-phase
load flow', Proc. IEE, 121 (7), 630.
5. K. A. Birt, J. J. Graffy, J. D. McDonald and A. H. El-Abiad, 1976, 'Three-phase load-
flow program', Trans. IEEE, PAS-95, 59.
6. J. Arrillaga and B. J. Harker, 1978. 'Fast decoupled three-phase load flow', Proc. IEE,
125 (8), 734-740.
7. B. Stott, Load flows for a.c. and a.c.-d.c. systems.' Ph.D. thesis, University of
Manchester, 1971.
8. B. Stott and 0. Alsac, 'Fast decoupled load flow', Trans. IEEE PAS-93, 1978, 859.
9. K. Zollenkopf, 1970. 'Bifactorization-basic computational algorithm and pro-
gramming techniques', Conference on Large Sets of Sparse Linear Equations, Oxford,
75-96.
 A. C.-D. C. Load Flow
6.1. INTRODUCTION
H.V.D.C. transmission is now an acceptable alternative to a.c. and is proving an
economical solution not only for very long distance but also for underground and
submarine transmission as well as a means of interconnecting systems ofdifferent
frequency or with problems of stability or fault level.
The growing number of schemes in existence and under consideration
demands corresponding modelling facilities for planning and operational
purposes.
The basic load flow has to be substantially modified to be capable of modelling
the operating state of the combined a.c. and d.c. systems under the specified
conditions of load, generation and d.c. system control strategies.
Having established the superiority of the fast decoupled a.c. load flow") the
integration of h.v.d.c. transmission is now described with reference to such
an algorithm. This is carried out using sequential and unified approaches.
In the sequential a p p r o a ~ h ( ~ )the
, ( ~ )a.c. and d.c. equations are solved
separately and thus the integration into existing load-flow programs is carried out
without significant modification or restructuring of the a.c. solution technique.
For the a.c. iterations each converter is modelled simply by the equivalent real or
reactive power injection at the terminal busbar. The terminal busbar voltages
obtained from the a.c. iteration are then used to solve the d.c. equations and
consequently, new power injections are obtained. This process continues
iteratively to convergence.
Alternatively, the more sophisticated unified m e t h o d ~ ( ~ ) , ( give
~),(~full
) re-
cognition to the interdependence between a.c. and d . ~ system . equations and
simultaneously solve the complete set of equations.

6.2. FORMULATION O F THE PROBLEM

The operating state of the combined power system is defined by the vector

where
P is a vector of the voltage magnitudes at all a.c. system busbars.
 0 is a vector of the angles at all a.c. system busbars (except the reference bus
which is assigned 8 = 0).
2 is a vector of d.c. variables.
Chapter 4 has described the use of P and 8 as a.c. system variables and the
selection of d.c. variables 2 is discussed in Section 6.3.
The development of a Newton-Raphson based algorithm requires the
formulation of n independent equations in terms of the n variables.
The equations which relate to the a.c. system variables are derived from the
specified a.c. system operating conditions. The only modification required to the
usual real and reactive power mismatches occurs for those equations which relate
to the converter terminal busbars. These equations become:

where
Pterm(ac)is the injected power at the terminal busbar as a function of the a.c.
system variables
Pterm(dc)is the injected power at the terminal busbar as a function of the d.c.
system variables
c g m is the usual a.c. system load at the busbar.
and similarly for Qterm(dc)and Qterm(ac)
The injected powers Qterm(dc)and Pterm(dc)are functions of the convertor a.c.
terminal busbar voltage and of the d.c. system variables, i.e.

The equations derived from the specified a.c. system conditions may therefore be
summarized as :

where the mismatches at the converter terminal busbars are indicated separately.
A further set of independent equations are derived from the d.c. system
conditions. These are designated,

for k = 1, number of converters present.

The d.c. system equations (6.2.3), (6.2.4) and (6.2.6) are made independent of the
a.c. system angles 0 by selecting a separate angle reference for the d.c. system
variables as defined in Fig. 6.2. This improves the algorithmic performance by
effectively decoupling the angle dependence of a.c. and d.c. systems.
The general a.c.-d.c. load flow problem may therefore be summarized as the
solution of:

where the subscript 'term' refers to the converter a.c. terminal busbar.

6.3. D.C. SYSTEM MODEL

The selection of variables 2 and formulation of the equations require several basic
assumptions which are generally accepted'') in the analysis of steady state d.c.
converter operation. These are:
(i) The three a.c. voltages at the terminal busbar are balanced and sinusoidal.
(ii) The converter operation is perfectly balanced.
(iii) The direct current and voltage are smooth.
(iv) The converter transformer is lossless and the magnetizing admittance is
ignored.

Converter variables
Under balanced conditions similar converter bridges attached to the same a.c.
terminal busbar will operate identically regardless of the transformer connection.
They may therefore be replaced by an equivalent single bridge for the purpose of
single-phase load flow analysis. With reference to Fig. 6.1 the set of variables
illustrated, representing fundamental frequency or d.c. quantities permits a full
description of the converter system operation.
An equivalent circuit for the converter is shown in Fig. 6.2 which includes the
modification explained in Section 6.2 as regards the position of angle reference.
The variables, defined with reference to Fig. 6.2 are:
Vterm& converter terminal busbar nodal voltage (phase angle referred to
converter reference)

-
Is le"

I-
I
IP
--t
- -
I

Fig. 6.1. Basic d.c. convertor (angles refer to a.c. system reference)
 Fig. 6.2. Single-phase equivalent circuit for basic converter. (Angles
referred to d.c. reference)

E fundamental frequency component of the voltage waveform at the

converter transformer secondary
I,, I, fundamental frequency component of the current waveshape on the
primary and secondary of the converter transformer respectively
a firing delay angle
a transformer off-nominal tap ratio
Vd average d.c. voltage
Id converter direct current
These ten variables, nine associated with the converter, plus the a.c. terminal
voltage magnitude V,,,,, form a possible choice of 3 for the formulation of
equations (6.2.3), (6.2.4) and (6.2.6).
The minimum number of variables required to define the opcration of the
system is the number of independent variables. Any other system variable or
parameter (e.g. Pdcand Qdc)may be written in terms of these variables.
Two independent variables are sufficient to model a d.c. converter, operating
under balanced conditions, from a known terminal voltage source. However, the
control requirements of h.v.d.c. converters are such that a range of variables, or
functions of them (e.g. constant power), are the specified conditions. If the
minimum number of variables are used, then the control specifications must be
translated into equations in terms of these two variables. These equations will
often contain complex non-linearities, and present difficulties in their derivation
and program implementation. In addition, the expressions used for Pdcand Q,, in
equations (6.2.1.) and (6.2.2.) may be rather complex and this will make the
programming of a unified solution more difficult.
For these reasons, a nonminimal set of variables is recommended, i.e. all
variables which are influenced by control action are retained in the model. This is
in contrast to a.c. load flows where, due to the restricted nature of control
specifications, the minimum set is normally used.
The following set of variables permits simple relationships for all the normal
control strategies.
[%I= Vd, Id, a, cos a, 41
Variable 4 is included to ensure a simple expression for Qdc.While this is
important in the formulation of the unified solution, variable 4 may be omitted
with the sequential solution as it is not involved in the formulation of any control
specification; cos a is used as a variable rather than a to linearize the equations
and thus improve convergence.

D.C. per unit system

To avoid per unit to actual value translations and to enable the use of comparable
convergence tolerances for both a.c. and d.c. system mismatches, a per unit
system is also used for the d.c. quantities.
Computational simplicity is achieved by using common power and voltage
base parameters on both sides of the converter, i.e. the a.c. and d.c. sides.
Consequently, in order to preserve consistency of power in per unit, the direct
current base, obtained from (MVAB)/VB,has to be fi times larger than the a.c.
current base.
This has the effect of changing the coefficients involved in the a.c.-d.c. current
relationships. For a perfectly smooth direct current and neglecting the com-
mutation overlap, the r.m.s. fundamental components of the phase current is
related to Idby the approximation,

Translating equation (6.3.1) to per unit yields:

I,(p.u.) f i J~.I,(~.U.)
= -.
n
and if commutation overlap is taken into account, as described in Chapter 3, this
equation becomes

where k is very close to unity. In load-flow studies, equation (6.3.2) can be made
sufficiently accurate in most cases by letting:
k = 0.995

Derivation of equations
The following relationships are derived for the variables defined in Fig. 6.2. The
equations are in per unit.
(i) The fundamental current magnitude on the converter side is related to the
direct current by the equation

(ii) The fundamental current magnitudes on both sides of the lossless

 transformer are related by the off-nominal tap, i.e.

(iii) The d.c. voltage may be expressed in terms of the a.c. source commutating
voltage referred to the transformer secondary, i.e.

The converter a.c. source commutating voltage is the busbar voltage on

the system side of the converter transformer, Vlerm.
(iv) The d.c. current and voltage are related by the d.c. system configuration,

e.g. for a simple rectifier supplying a passive load,

(v) The assumptions listed at the beginning of this section prevent any real
power of harmonic frequencies at the primary and secondary busbars.
Therefore, the real power equation relates the d.c. power to the
transformer secondary power in terms of fundamental components only,
i.e.
VdeId= E.I;cos $ (6.3.7)
(vi) As the transformer is lossless, the primary real power may also be equated
to the d.c. power, i.e.

(vii) The fundamental component of current flow across the converter

transformer can be expressed as
sin 4
I, = B;sin $ - B;a- VIerm (6.3.9)
where jB, is the transformer leakage susceptance.
So far, a total of seven equations have been derived and no other independent
equation may be written relating the total set of nine converter variables.
Variables, I,, I,, E and $ can be eliminated as they play no part in defining
control specifications.
Thus equations (6.3.3), (6.3.4), (6.3.7) and (6.3.8) can be combined into
Vd - kl .a. Vlermcos4 = 0 (6.3.10)
where k1 = k(3 f i l n ) .
The final two independent equations required are derived from the specified
control mode.
The .d.c. model may thus be summarized as follows:
130
where

R(4) = control equation

R(5) = control equation
and
2 = [ V,, I,, a, cos a, 41
Vtermcan either be a specified quantity or an a.c. system variable. The equations
for P,, and Q,, may now be written as,

Incorporation of control equations

Each additional converter in the d.c. system contributes two independent
variables to the system and thus two further constraint equations must be derived
from the control strategy of the system to define the operating state. For example,
a classical two-terminal d.c. link has two converters and therefore requires four
control equations. The four equations must be written in terms of the ten d.c.
variables (five for each converter).
Any function of the ten d.c. system variables is a valid (mathematically) control
equation so long as each equation is independent of all other equations. In
practice, there are restrictions limiting the number of alternatives. Some control
strategies refer to the characteristics of power transmission (e.g. constant power
or constant current), others introduce constraints such as minimum delay or
extinction angles.
Examples of valid control specifications are;
(i) Specified converter transformer tap,
a- -
-0
(ii) Specified d.c. voltage
v, - ViP = 0
 (iii) Specified d.c. current
I, - IS,p= 0
(iv) Specified minimum firing angle
COS a - COS amin
=0

(v) Specified d.c. power transmission

These control equations are simple and are easily incorporated into the solution
algorithm. In addition to the usual control modes, nonstandard modes such as
specified a.c. terminal voltage may also be included as converter control
equations (see Section 6.5).
During the iterative solution procedure the uncontrolled converter variables
may go outside prespecified limits. When this occurs, the offending variable is
usually held to its limit value and an appropriate control variable is freedj4)

Inverter operation
All the equations presented so far are equally applicable to inverter operation.
However, during inversion it is the extinction advance angle (y) which is the
subject of control action and not the firing angle (a). For convenience therefore,
equation R(2) of (6.3.11) may be rewritten as

This equation is valid for rectification or inversion. Under inversion, V,, as

calculated by equation (6.3.15), will be negative.
To specify operation with constant extinction angle the following equation is
used
cos(.n - y) - cos(n - ysP)= O
where ysP is usually y minimum for minimum reactive power consumption of the
inverter.

6.4. SOLUTION TECHNIQUES

The converter model developed in Section 6.3 is to be incorporated with a fast
decoupled a.c. load flow algorithm with the minimum possible modification of
the latter in order to retain its computational advantages.
The solution is first discussed with reference to a single converter connected to
an a.c. busbar. The extension to multiple or multiterminal d.c. systems is
relatively trivial and is discussed in Section 6.6.

Unified solution
The unified method gives recognition to the interdependence of a.c. and d.c.
system equations and simultaneously solves the complete system. Referring to
equation (6.2.7) the standard Newton-Raphson algorithm involves repeat
solutions of the matrix equation:

where J is the matrix of first order partial derivatives.

and

Applying the a.c. fast decoupled assumptions to all Jacobian elements related
to the a.c. system equations, yields:

where all matrix elements are zero unless otherwise indicated. The matrices [B']
and [B"] are the usual single-phase fast decoupled Jacobians and are constant in
value. The other matrices indicated vary at each iteration in the solution process.
A modification is required for the element indicated as B:i in equation (6.4.6).
This element is a function of the system variables and therefore varies at each
iteration.
The use of an independent angle reference for the d.c. equations results in

i.e. the diagonal Jacobian element for the real power mismatch at the converter
terminal busbar depends on the a.c. equations only and is therefore the usual fast
decoupled B' element.
In addition,
aR/a&erm =0

which will help the subsequent decoupling of the equation.

 In order to maintain the block successive iteration sequence of the usual fast
decoupled a.c. load flow, it is necessary to decouple equation (6.4.6).Therefore the
Jacobian submatrices must be examined in more detail. The Jacobian sub-
matrices are:

Following decoupled load-flow practice

1
=0 + -(ap,ermw~laV,erm)
Vterm
and since
Pterm(dc)= Vd.ld

therefore: DD = 0
Similarly
1
CAA'I = - - [ a ~ ~ ~ ~ ~ ~ l a x l
Vterm

1
[AA"] = -[aAQterm/a2]
Vterm

1
B!! = -CaAQtermla Vterml
" Vterm

In the above formulation the d.c. variables 2 are coupled to both the real and
reactive power a.c. mismatches. However, equation (6.4.6) may be separated to
enable a block successive iteration scheme to be used.
134
The d.c. mismatches and variables can be appended to the two fast decoupled
a.c. equations in which case the following two equations result

The iteration scheme illustrated in Fig. 6.3 is referred to as-PDC, QDC-and

the significance of the mnemonic should be obvious.
The algorithm may be further simplified by recognizing the following physical
characteristics of the a.c. and d.c. systems;
- The coupling between d.c. variables and the a.c. terminal voltage is strong.
- There is no coupling between d.c. mismatches and a.c. system angles.
- Under all practical control strategies the d.c. power is well constrained and
this implies that the changes in d.c. variables 2 do not greatly affect the real
power mismatches at the terminals. This coupling, embodied in matrix AA' of
equation (6.4.7) can therefore be justifiably removed.
These features justify the removal of the d.c. equations from equation (6.4.7) to
yield a -P, QDC-block successive iteration scheme represented by the following
two equations,
[ A P / V ] = [B'] [A81

Bii AA"

Programming considerations for the unified algorithm

In order to retain the efficiency of the fast decoupled load flow, the B' and B"
matrices must be factorized only once before the iterative process begins.
 I DATAvINPUT 1
Triangulate B' and 6"
convertor busbars not used as pivots

I calculate AP (0.c. system only 1 ]

t

.
I ~ a l c u l a t ePkm (dc) and d r residuals ( 5 1 I

I Mrm d.c. Jacobin

w
Matrix

I
1

I Forward reduction of Vector A ~ / V ]

*
I Calculate A d ( a . c . svstem onlv
+
I Calculate qerm
(d.c.1 and d c residuals ( #1
I [sropb
Converge
*
I Form d.c. Jacobian Matrix
w

I Forward reduction of vector A

&
d /V 1 1
[solve reduced quatidn for Vterm and A 2 I
t
I Back substitute for A P 1
I Update Fond I

Fig. 6.3. Flow chart for unified single-phase a.c.-d.c. load flow

The Jacobian elements related to the d.c. variables are non-constant and must
be re-evaluated at each iteration. It is therefore necessary to separate the constant
and nonconstant parts of the equations for the solution routine.
Initially, the a.c. fast decoupled equations are formed with the d.c. link ignored
(except for the minor addition of the filter reactance at the appropriate a.c.
busbar). The reactive power mismatch equation for the a.c. system is:
136
where
AQIerm = Q:Lm - Qterm(ac).
is the mismatch calculated in the absence of
the d.c. converter.
and
B" is the usual constant a.c. fast decoupled Jacobian.
After triangulation down to, but excluding the busbars to which d.c. converters
are attached, equation (6.4.11) becomes

where (AQ/ V)" and (AQterm/ Vterm)"signify that the left-hand side vector has been
processed and matrix B"' is the new matrix B" after triangulation.
This triangulation (performed before the iterative process) may be achieved
simply by inhibiting the terminal busbars being used as pivots during the optimal
ordering process.
The processing of AQ indicated in the equation is actually performed by the
standard forward reduction process used at each iteration.
The d.c. converter equations may then be combined with equation (6.4.12)as
follows :
AV

0 BI;+BL(dc) AA" A'term

A2

where

The unprocessed section, i.e.

may then be solved by any method suitable for nonsymmetric matrices.

 The values of A x and AVtermare obtained from this equation and AVt,,, is then
used in a back substitution process for the remaining A P to be completed, i.e.
equation (6.4.12) is solved for A r.
The most efficient technique for solving equation (6.4.14) depends on the
number of converters. For six converters or more, the use of sparsity storage and
solution techniques are justified, otherwise all elements should be stored. The
method suggested here is a modifed form of gaussian elimination where all
elements are stored but only nonzero elements processed.'')

Sequential method
The sequential method results from a further simplification ofthe unified method,
i.e. the a.c. system equations are solved with the d.c. system modelled simply as a
real and reactive power injection at the appropriate terminal busbar. For a d.c.

(all ax. busbars)

1

[~olculateAP (total system ) and d.c. residuals A ]

-
onverge
OY- --7 -
[!Solve equation 6.415and update ( ill
.1.
a
]Calculate Ad (total system) and d.c residuals

TForm d.c. Jocobian matrix

Fig. 6.4. Flow chart for sequential single-phase a.c.-d.c. load

flow
solution the a.c. system is modelled simply as a constant voltage at the converter
a.c. terminal busbar.
The following three equations are solved iteratively to convergence.
[AF/P] = [B'] CAB]
[AQ/V] = [B"] [A v]
[R] = [A] [AX]
This iteration sequence, referred to as P, Q, DC, is illustrated in the flow chart of
Fig. 6.4 and may be summarized as follows;
(i) Calculate AF/P, solve equation (6.4.15) and update 8.
(ii) Calculate AQIV, solve equation (6.4.16) and update R
(iii) Calculate d.c. residuals, R, solve equation (6.4.17) and update 5.
(iv) Return to (i).
With the sequential method the d.c. equations need not be solved for the entire
iterative process. Once the d.c. residuals have converged, the d.c. system may be
modelled simply as fixed real and reactive power injections at the appropriate
converter terminal busbar. The d.c. residuals must still be checked after each a.c.
iteration to ensure that the d.c. system remains converged.
However, in order to establish a direct comparison between the unified and
sequential algorithms the d.c. equations are continued to be solved until both a.c.
and d.c. systems have converged. This ensures that the sequential technique is an
exact parallel of the corresponding unified algorithm.
Alternatively, the d.c. equations can be solved after each real power as well as
after each reactive power iteration and the resulting sequence is referred to as
P, DC, Q, DC. As in the previous method, the d.c. equations are solved until all
mismatches are within tolerance.

6.5. CONTROL OF CONVERTER A.C. TERMINAL VOLTAGE

A converter terminal voltage may be specified in two ways:
(a) By local reactive power injection at the terminal. In this case no reactive
power mismatch equation is necessary for that busbar and the relevant
variable (i.e. AVlerm)is effectively removed from the problem formulation.
This is the situation where the converter terminal busbar is a P - V busbar.
(b) The terminal voltage may be specified as a d.c. system constraint. That is,
the d.c. converter must absorb the correct amount of reactive power so that
the terminal voltage is maintained constant.
With the unified method, the equation

is written as one of the two control equations. This would lead to a zero row in
equation (6.4.7) and therefore during the solution of equation (6.4.7) some other
variable (e.g. tap ratio) must be specified instead. Although d.c. convergence is
marginally slower for the PDC, QDC iteration, the d.c. system is overconverged in
this iteration scheme and the overall convergence rate is practically unaffected.
With the sequential method equation (6.5.1)cannot be written. The terminal
busbar is specified as a P - V busbar and the control equation

is used, where QsLrn(dc) is taken as the reactive power required to maintain the
voltage constant. The specified reactive power thus varies at each iteration and
this discontinuity slows the overall convergence.

6.6. EXTENSlON T O MULTIPLE AND/OR MULTITERMINAL D.C.

SYSTEMS
The basic algorithm has been developed in previous sections for a single d.c.
converter. Each additional converter adds a further five d.c. variables and a
corresponding set of five equations. The number of a.c. system Jacobian elements
which become modified in the unified solutions is equal to the number of
converters.
As an example, consider the system shown in Fig. 6.5. The system represents
the North and South ~sl'andsof the New Zealand Electricity Division, 220 kV a.c.
system. At present converters 1,2 and 3 are in operation. Converters 1 and 2 form
the 600 MW, 500 kV d.c. link between the two islands. Converter 3 represents a
420 MW aluminium smelter. A further t hree-terminal d.c. interconnection has
been added. (Converters 4, 5 and 6 ) to illustrate the flexibility of the algorithm.
Normally, Converter 4 will operate in the rectifier mode with converters 5 and
6 in the inversion mode.
The reactive power-d.c. Jacobian for the unified method has the following
structure,

%+
82-Bus

t
North Island
system

Fig. 6.5. Multiterminal d.c. system

where FAoDis the part of B which becomes modified. Only the diagonal elements
become modified by the presence of the converters.
Off diagonal elements will be present in BiODif there is any a.c. connection
between converter terminal busbars. All off diagonal elements of B B and AA"
are zero.
In addition, matrix A is block diagonal in 5 x 5 blocks with the exception of the
d.c. interconnection equations.
Equation R(3) of (6.3.11) in each set of d.c. equations is derived from the d.c.
interconnection. For the six-converter system shown in Fig. 6.5 the following
equations are applicable.

This example indicates the ease of extension to the multiple converter case.

6.7. D.C. CONVERGENCE TOLERANCE

The d.c. p.u. system is based upon the same power base as the a.c. system and on
the nominal open circuit a.c. voltage at the converter transformer secondary. The
p.u. tolerances for d.c. powers, voltages and currents are therefore comparable
with those adopted in the a.c. system.
In general, the control equations are of the form
X-~P=O
where X may be the tap or cosine of the firing angle, i.e. they are linear and are
thus soived in one d.c. iteration. The question of an appropriate tolerance for
these mismatches is therefore irrelevant.
An acceptable tolerance for the d.c. residuals which is compatible with the a.c.
system tolerance is typically 0.001 p.u. on a 100 MVA base, i.e. the same as that
normally adopted for the a.c. system.

6.8. TEST SYSTEM AND RESULTS

The A.E.P. standard 14-bus test system is used to show the convergence
properties of the a.c.-d.c. algorithms, with the a.c. transmission line between
busbars 5 and 4 replaced by a h.v.d.c. link. As these two buses are not voltage
controlled, the interaction between the a.c. and d.c. systems will therefore be
considerable.
 Table 6.1. Convergence results

Case specification Number of iterations to convergence (0.1 MW/MVAR)

Sequential methods
Specified d.c. Unified methods
link constraints (5 variables) (5 variables) (4 variables)
m-rectifier end
n-inverter end ~ P D C QDC
. 2 ~QDC
. ~P,Q.DC~P.DC,Q.DC ~P.Q.DC~P.DC,Q.DC

am ' d m y n ' d n 493

a; P d m a n Vdn Failed
a m P d m a n Vdn Failed
a m P d m Y n Vdn 4,4
am Pdm yn an 4,4
am P, a m Yn 4,3
am I d y n 'dm 4,3
am ' d m y n Pdn 4,4
Case.! with initial
condltlon errors
9 50 per cent error 4,3 4,3 4,4 4,3 4,4 4,3
10 80 per cent error 5,4* 6,5* 7,6* 5,4* 4.4 4,3

Where indicates a false solution.

Table 6.2. Characteristics of d.c. link

Converter I Converter 2

A.c. Busbar Bus 5 Bus 4

D.c. voltage base 100 kV 100 kV
Transformer reactance 0.126 0.0728
Commutation reactance 0.126 0.0728
Filter admittance B; 0.478 0.629

D.c. link resistance 0.334 ohms

Control parameters for Case 1

D.c. link power 58.6 MW -

Rectifier firing angle (deg) 7 -

Inverter extinction angle (deg) -- 10
Inverter d.c. voltage - - 128.87 kV
Filters are connected to a.c. terminal busbar.
Note: All reactances are in p.u. on a 100 MVA base.

Various control strategies have been applied to the link and the convergence
results for the various algorithms are given in Table 6.1. The number of iterations
(i,j) should be interpreted as follows:
- i is the number of reactive power-voltage updates required.
-j is the number of real power-angle updates.
Although the number of d.c. iterations varies for the different sequences, this is of
secondary importance and may if required be assessed in each case from the
number 0fa.c. iterations. In this respect, a unified QDC iteration is equivalent to a
Q iteration and a DC iteration executed separately.
The d.c. link data and specified controls for Case 1 are given in Table 6.2 and
the corresponding d.c. link operation is illustrated in Fig. 6.6. The specified
conditions for all cases are derived from the results of Case 1. Under those
conditions, the a.c. system in isolation, (with each converter terminal modelled as
an equivalent a.c. load) requires (4, 3) iterations. The d.c. system in isolation
(operating from fixed terminal voltages) requires two iterations under all control
strategies.

Bus 5 Bus 4

All angles are in degrees. D.C.voltages and current are in kVand Amp respectively.
D.C.resistance is in ohms. AC.powers (P,Q)are in MW and MVARs.
Fig. 6.6. D.c. link operation for Case I

Unified cases
The results in Table 6.1 show that the unified methods provide fast and reliable
convergence in all cases.
For the unified methods 1 and 2 the number of iterations did not exceed the
number required for the a.c. system alone.

Sequential cases
The sequential method (P,Q,DC) produces fast and reliable convergence
although the reactive power convergence is slower than for the a.c. system alone.
With the removal of the variable 4, Q,,,, (dc) converges faster but the
convergence pattern is more oscillatory and an overall deterioration of a.c.
voltage convergence results.
With the second sequential method, (P, DC, Q, DC) convergence is good in all
cases except 2 and 3, i.e. the cases where the transformer tap and d.c. voltage are
specified at the inverter end. However, this set of specifications is not likely to
occur in practice.

Initial conditions for d.c. system

Initial values for the d.c. variables 2 are assigned from estimates for the d.c. power
and d.c. voltage and assuming a power factor of 0.9 at the converter terminal
busbar. The terminal busbar voltage is set at 1.0 p.u. unless it is a voltage
controlled busbar.
 This procedure gives adequate initial conditions in all practical cases as good
estimates of P,,,, (dc) and Vd are normally obtainable.
With starting values for d.c. real and reactive powers within f 50 per cent,
which are available in all practical situations, all algorithms converged rapidly
and reliably (see Case 9).

Effect of a.c. system strength

In order to investigate the performance of the algorithms with a weak a.c. system,
the test system described earlier is modified by the addition of two a.c. lines as
shown in Fig. 6.7.
Bus 5 Bus 4
I
4

Q Filters

Fig. 6.7. D.c. link operating from weak a.c. system

0
Filters

Table 6.3. Numbers of iterations for weak a.c. systems

Case specification x, = 0.3 x, = 0.4

Sequential Sequential
P.Q,DC P,Q,DC
m-rectifier Unified Unified
n-inverter P,QDC (i) (ii) P,QDC (1) (ii)

494 4,4 4,4 4,4 594 4,4

4,4 9,8 10,12 4,4 > 30 Diverges
4,3 9,8 10,12 4,3 > 30 Diverges
4,3 65 7,7 4,3 28,27 > 30

(i) using the five variable formulation; (ii) using the four variable formulation.

The reactive power compensation of the filters was adjusted to give similar d.c.
operating conditions as previously.
The number ofiterations to convergence for the most promising algorithms are
shown in Table 6.3 for the control specifications corresponding to Cases 1 to 4 in
the previous results.
The different nature of the sequential and unified algorithms is clearly
demonstrated. The effect of the type of converter control is also shown. For Case
11, both the d.c. real power and the d.c. reactive powers are well constrained by
the converter control strategy. Convergence is rapid and reliable for all methods.
In all other cases, where the control angle at one or both converters is free, an
oscillatory relationship between converter a.c. terminal voltage and the reactive
power of the converter is possible. This leads to poor performance of the
sequential algorithms.
To illustrate the nature of the iteration, the convergence pattern of the
converter reactive power demand and the a.c. system terminal voltage of the
rectifier is plotted in Fig. 6.8.
A measure of the strength of a system in a load-flow sense is the short circuit to
converter power ratio (SCR) calculated with all machine reactances set to zero.
This short-circuit ratio is invariably much higher than the usual value.
In practice, converter operation has been considered down to a SCR of 3. A
survey ofexisting schemes shows that, almost invariably, with systems of very low
SCR, some form of voltage control, often synchronous condensers, is an integral
part of the converter installation. These schemes are therefore often very strong in
a load-flow sense.
It may therefore be concluded that the sequential integration should converge
in all practical situations although the convergence may become slow if the
system is weak in a load-flow sense.

Fig. 6.8. Convergence pattern for a.c.-d.c. load flow with weak a.c.
system. (i) Sequential method (P,Q, DC five variable); (ii) Unified
method (P, QDC)
Discussion of convergence properties
The overall convergence rate of the a.c.-d.c. algorithms depends on the successful
interaction of the two distinct parts. The a.c. system equations are solved using
the well-behaved constant tangent fast decoupled algorithm, whereas the d.c.
system equations are solved using the more powerful, but somewhat more erratic,
full Newton-Raphson approach.
The powerful convergence of the Newton-Raphson process for the d.c.
equations can cause overall convergence difficulties. If the first d.c. iteration
occurs before the reactive power-voltage update then the d.c. variables are
converged to be compatible with the incorrect terminal voltage. This introduces
an unnecessary discontinuity which may lead to convergence difficulties in the
sequential method. In the unified approach the powerful convergence of the d.c.
equations is dampened by the reflection of the a.c. mismatches onto the changes
in d.c. variables. This gives faster and better behaved convergence. The solution
time of the d.c. equations is normally small compared to the solution time of the
a.c. equations. The relative efficiencies of the alternative algorithms may there-
fore be assessed by comparing the total numbers of voltage and angle updates.
In general, those schemes which acknowledge the fact that the d.c. variables are
strongly related to the terminal voltage give the fastest and most reliable
performance. The unified methods are the more reliable and the P, QDC solution
is the most efficient. Of the sequential methods, the (P,Q, DC) solution is only
marginally inferior to the unified method.
When the busbar to which the converters are attached is voltage controlled (as
is often the case) the two approaches become virtually identical as the interaction
between a.c. and d.c. systems is much smaller.
The only computational difference between a unified and a comparable
sequential iteration is that the d.c. Jacobian equation for the unified method
(equation 6.4.14) is slightly larger. The difference is one additional row and
column for each converter present. In terms of computational cost per iteration
the corresponding unified and sequential algorithms are virtually identical.
(i) In cases where the a.c. system is strong, both the unified and sequential
algorithms may be programmed to give fast and reliable convergence.
(ii) If the a.c. system is weak the sequential algorithm is susceptible to
convergence problems. Thus, in general, the unified. method is recom-
mended due to its greater reliability.

6.9. NUMERICAL EXAMPLE

The complete New Zealand Primary transmission system described in Appen-
dix I was used as a basis for a planning study which included an extra
multiterminal h.v.d.c. scheme, i.e. involving six converter stations as illustrated in
Fig. 6.5.
Representative input and output information obtained from the computer is
given in the following pages.
 I:
3
........
ooococooc
I% oaooo3ooz
.-,a
Z S 30000330C
-z 0 3
I m
Z OOOOOCOOC
m Q. za 49003q03=
W W L L 3) o........
5 x 0 UT
C
*93000000
md
I I
3 3 0
Z Z Z
a LU
t-
z
Lu3
.........
OCCCOCCOCc
OCCCCCOCOC
 " 0 0 0 0
a 0 0 0 0
u o o o o

-
CI
1 0 0 0 0
0 . 0 0 0 0
C l o o o o
u o o o o

-
C(
1 0 0 0 0
m o o 0 0
C l o o o o

m
.....
u o o

o o
o

0
o

0
-
rn
C(
1 0 0 0 0
P \ o o o o
a 0 0 0 0
.....
u o o o o

-.
C(
1 0 0 0 0
9 0 0 0 0
G O O O O
.....
u o o o o

C( I
1 0 0 0 0
Y l o o o o
a 0 0 0 0
~ 0 0 0 0

-
C(
1 0 0 0 0
- 0 0 0 0
G O O O O
.....
0
& O O O O
000000000m000000000
L 000000000ln000000000 - 0 0 0 0
a
i- ...................
ooooooooooooooooooo
"dddd-ld-l"""-ld""""d"
G
..
1 0 0 .
"
0 0
m o - l o o
C l o o o o
u
.....o o

P Y 0 0 6 0
o o

n
C(
1 0 0 0 0
r.4
0 0 0 0
U O O O O
.....
C L O O O
r 4 0 0 0 0
O

a
SOLUTION C O N V E R G E D IN 7 P - D AND 6 n-v ITERATIONS

LOAD GENERATION AC LOSSES HISHATCH SHUNTS

HW H VA H nu HVAH nu HVAR nu HVAR HVAR
4496.80 1518.60 5326.58 791.80 194.91 -306.06 534.87 -182.89 37.85

CONVERTER I S OPERATING I N THE I N V E R T I O N MODE

THE CONTROL ANGLE I S THE E X T I N C T I O N ADVANCE ANGLE
DC POWER SUPPLIED TO THE AC SYSTEM = 200.00 HW

CONVERTER AC VOLTAGE TRANSFORMER TAP CONTROL ANGLE COHMUTATION ANGLE DC CURRENT DC VOLTAGE
(K-VOLTS) (PER CENT) (DEGS) (DEGS) (K-AMPS) (K-VOLTS)
97.08 -5.46 8.00 19.93 0.81 -245.67

POWER TRANSFERS
L I N K TERMINAL POWER = -200.00 HW 110.89 HVAR
FROM TRANSFORMER TO CONVERTER = - 2 0 0 . 0 0 HW 74.63 HVAR
REACTIVE POWER OF F I L T E R S = 120.83 HVAR

BUSBAR DATA L I N E AND TRANSFORMER DATA

GENERATION LOAD SHUNT
BUS NAME VOLTS ANGLE HW HVAR HW HVAR HVAR BUS NAME HU HVAR

104 AVIEHORE-220 1.052 4.78 220.00 -33.87 0.00 0.00

HISHATCH 0.000 0.000

0.00 ...................................
1 0 4 AVIEHORE-220 -41.83 7.88
104 AVIEHORE-220 -41.83 7.88
255 TUIZEL---220
MISMATCH
26.47
500.000
6.87
-110.672 -
-.--
167 ISLINGTON220 -120.18 -76.16 a
1 8 1 LAND-T02-220 -209.40 -19.64
MISMATCH -0.019 -0.002
 CONVERTER ATlAChEO TO b U S NUHBEH 7

NOMINAL DC VOLTAGE NOMINaL DC VOLTAGE

MAXIMUM DC VOLTAGE M A X I W (IC VOLTAGE
MINUULY OC VOLTAbE M I N I M U M DC VOLTAGE
MAXIMUM DC CURRENT MAXIMUM D C CUPRENT

COWIFTATION R E A C T L N E (P.U. )

TRANSFORMLQ REACTANCE (PoU. )

. F I R I N G ANGLE:MINIWUM(D€G)
MAXICUM(DEG)
TRANSFORMER TAP:MINIMUM(P.C.)
MAXIMUM ( P o c o )
INCRIiMENT
F l L T l i R REACTANCE(P.U.)
NUMBER OF BRIDGES I N SERIES*
 15 1
6.10 REFERENCES
1. B. Scott and 0.Alsac, 1974.'Fast decoupled load flow', Trans. IEEE, PAS-93 (3), 859-
869.
2. H. Sato and J. Arrillaga, 1969. 'Improved load-flow techniques for integrated a.c.-d.c.
systems', Proc. IEE, 116 (4), 525-532.
3. J. Reeve, G. Fahmy, and B. Stott, 1976. 'Versatile load-flow method for multi-terminal
h.v.d.c. systems', Paper F76-354-1. IEEE PES Summer Meeting, Portland,
4. B. Stott, 'Load flow for a.c. and integrated a.c.-d.c. systems.' Ph.D. Thesis, University of
Manchester, 1971.
5. D. A. Braunayal, L. A. h a f t , and J. L. Whysong, 1976. 'Inclusion of the converter and
transmission equations directly in a Newton power flow', IEEE Trans. PAS-75 (I), 76-
88.
6. J. Arrillaga, B. J. Harker, K. S. and Turner, 1980. 'Clarifying an ambiguity in recent
a.c.-d.c. load-flow formulations', Proc. IEE, 127, Pt. C (5), 324-325.
7. P. Bodger, 1977. 'Fast decoupled a.c. and a.c.-d.c. load flows.' Ph.D. Thesis, University
of Canterbury, New Zealand.
 Three-Phase A.C. -D.C. Load
Flow
7.1. INTRODUCTION
Any converter which is operatin g from an unbala.nced a.c. system will itself
operate with unbalanced power flows and unsymmetric valve conduction
periods. In addition any unbalance present in the converter control equipment or
any asymmetry in the converter transformer will introduce additional unbalance.
Considerable interaction exists between the unbalanced operation of the a.c.
and d.c. systems. The exact nature of this interaction depends on features such as
the converter transformer connection and the converter firing controller.
High-power converters often operate in systems of relatively low short-circuit
ratios where unbalance effects are more likely to be significant and require
additional consideration. The steady-state unbalance and its effect in converter
harmonic current generation may also influence the need for transmission line
transpositions and the means of reactive power compensation.
The converter model for unbalanced analysis is considerably more complex
than those developed for the balanced case in Chapters 3 and 6. The additional
complexity arises from the need to include the effect of the three-phase converter
transformer connection and of the different converter firing control modes. Early
h.v.d.c. control schemes were based on phase angle control, where the firing of
each valve is timed individually with respect to the appropriate crossing of the
phase voltages. This control scheme has proved susceptible to harmonic stability
problems when operating from weak a.c. systems. An alternative control, based
on equidistant firings on the steady state, is generally accepted to provide more
stable operation, ( 1 ) . ( 2 ) ~ ( 3 ) Under normal steady-state and perfectly-balanced
operating conditions, there is no difference between these two basic control
strategies. However, their effect on the a.c. system and d.c. voltage and current
waveshapes during normal, but not balanced, operation, is quite different. A
three-phase converter model must be capable of representing the alternative
control strategies.
Similar techniques are available for the integration of the three-phase
converter model into the load-flow analysis as were discussed with respect to the
balanced single-phase analysis. Based upon the extensive investigations into the
behaviour of single-phase a.c.-d.c. load flow described in Chapter 6, the
sequential approach is considered the most appropriate for the integration of the
d.c. model into the three-phase load flow. The complexity of the unified approach
is not considered justified in the three-phase case because, in cases of difficult
convergence such as those involving very weak a.c. systems, it is possible to use
starting values derived from single-phase analysis.
This chapter describes the development of a model for the unbalanced
converter and its sequential integration with the three-phase fast decoupled load
flow described in Chapter 5.

7.2. FORMULATION O F THE THREE-PHASE A.C.-D.C. LOAD-FLOW

PROBLEM
The operating state of the combined system is defined by:

where :
are vectors of the balanced internal voltages at the generator
internal busbars;
pH are vectors of the three-phase voltages at every generator
terminal busbar and every load busbar;
-
x is a vector of the d.c. variables (as yet, undefined).
The significance of the three-phase a.c. variables was discussed in Chapter 5
and the selection of d.c. variables iis discussed in Section 7.3.
To enable a Newton-Raphson based technique to be used, it is necessary to
formulate a set of n independent equations in terms of the n variables describing
the system. As explained in Chapter 5, the equations which relate to the a.c.
system variables are derived from the specified a.c. system operating conditions.
The only modification to these equations, which results from the presence of the
d.c. system, occurs at the converter terminal busbars. These equations become:

where Pferm(dc)
and Q,Perm(dc)
are function of the a.c. terminal conditions and the
converter variables, i.e.
 The equations for the a.c. system may therefore be summarized as,

where the mismatches at the converter terminal busbars are indicated separately.
Further equations are derived from the d.c. system conditions.
That is, for each converter k a set of equations

is derived in terms of the terminal conditions and the converter variables 2.

Equations (7.2.3), (7.2.4) and (7.2.6) form a mathematical model of the d.c.
system suitable for inclusion into load-flow analysis.
The three-phase a.c.-d.c. load-flow problem may therefore be formulated as
the solution of,

for the set of variables (K 8 , ~ ) .

7.3. D.C.-SYSTEM MODELLING

The basic h.v.d.c. interconnection shown in Fig. 7.1 is used as a reference and its
extension to other configurations is clarified throughout the development of the
model. Under balanced conditions, the converter transformer modifies the source
voltages applied to the converter and also affects the phase distribution of current
and power. In addition, the a.c. system operation may be influenced (e.g. by a
zero-sequence current flow to a star-g-delta transformer) by the transformer
connection. Each bridge in Fig. 7.1 will thus operate with a different degree of
unbalance, due to the influence of the converter transformer connections, and
must be modelled independently. This feature is in contrast to the balanced d.c.
model where it is possible to combine bridges in series and in parallel into an
 4 Filters

Fig. 7.1. Basic h.v.d.c. interconnection

Filters

Primary Secondary

Fig. 7.2. Basic converter unit

equivalent single bridge. The dimensions of the three-phase d.c. model, will,
therefore, be much greater than the balanced d.c. model.
All converters, whether rectifying or inverting, are represented by the same
model (Fig. 7.2) and their equations are of the same form.

Basic assumptions
To enable the formulation of equation (7.2.6) and to simplify the selection of
variables 2 the following assumptions are made:
(i) The three a.c. phase voltages at the terminal busbar are sinusoidal.
(ii) The direct voltage and direct current are smooth.
(iii) The converter transformer is lossless and the magnetizing admittance is
ignored.
Assumptions (ii) and (iii) are equally valid for unbalanced three-phase analysis
as for single-phase analysis. Assumption (i) is commonly used in unbalanced
converter studies (4)*(5) and appears to be backed from the experience of existing
schemes. However, a general justification will require more critical examination
of the problem.
Under balanced operation only characteristic harmonics are produced and, as
filtering is normally provided at these frequencies, the level of harmonic voltages
will be small. However, under even small amounts of unbalance, significant
noncharacteristic harmonics may be produced and the voltage harmonic
distortion at the terminal busbars will increase. The possible influence of
harmonic voltages at the converter terminal, on the fundamental frequency
power flows to the converter, is considered in Section 7.7.

Selection of converter variables

The selection of converter variables was discussed in detail in Chapter 6, with
regard to the balanced converter model. The main considerations, also relevant
to the unbalanced three-phase converter model, were:

Phase1
Phase3 Phase 2

Fig. 7.3. Unbalanced converter vol-

tage and current waveform. (i) Phase
voltages; (ii) D.c. voltage waveform;
(iii) Assumed current waveshape for
Phase 1 (actual waveform is indicated
by dotted line)
 (i) For computing efficiency, the smallest number of variables should be used.
A minimum of six independent variables is required to define the operating
state of an unbalanced converter, e.g. the three firing angles and the three
transformer tap positions.
(ii) To enable the incorporation of a wide range of control specifications, all
variables involved in their formulation should be retained. The following
variables, defined with reference to Fig. 7.2 and 7.3, are required in the
formulation of the control specifications for unbalanced converter
operation :
a, Off-nominal tap ratios on the primary side.
1' U23/C2, U 2 1 1 C 3
Phase-to-phase source voltages for the converter referred to the
transformer secondary. C, are therefore the zero crossings for the timing
of firing pulses.
a, Firing delay angle measured from the respective zero crossing.
Vd Total average d.c. voltage from complete bridge.
I, Average d.c. current.
where i = 1,2 3 for the three phases involved.
In contrast to the balanced case, the secondary phase to phase source voltages
are included among the variables as they depend not only on the transformer taps
but also on the transformer connection. Moreover, the zero crossings, Ci are
explicitly required in the formulation of the symmetrical firing controller and
they are also included.
Although these fourteen variables do not constitute the final d.c. model it is
convenient to formulate equation (7.2.6) in terms of these variables at this stage,
i.e. vector 2 has the form:
C Ui, Ci, a,, ai, Vd, I d ]
The necessary fourteen equations are derived in the following section.

Converter angle reference

In the three-phase a.c. load flow described in Chapter 5 all angles are referred to
the slack generators internal busbar. Similarly to the single-phase a.c.-d.c. load
flow (Chapter 6) the angle reference for each converter may be arbitrarily
assigned. By using one of the converter angles (e.g. ,O,: in Fig. 7.2) as a reference,
the mathematical coupling between the a.c. system and converter equations is
weakened and the rate of convergence improved.

Per unit system

Similarly to the single-phase case, computational simplicity is achieved by using
common power and voltage bases on both sides of the converter.
In the three-phase case, however, the phase-neutral voltage is used as the base
parameter and therefore
MVA,,,, = Base Power per phase
Vba,, = Phase-neutral voltage base
The current base on the a.c. and d.c. sides are also equal. Therefore the p.u. system
does not change the form of any of the converter equations.

Converter source voltages

The phase-to-phase source voltages referred to the transformer secondary are
found by a consideration of the transformer connection and off-nominal turns
ratio. For example, consider the star-star transformer of Fig. 7.4.

7 Phase 1

1
Phase 2

Fig. 7.4. Star-Star transformer connection

The phase-to-phase source voltages referred to the secondary are:

which in terms of real and imaginary parts yield six equations.

D.C. voltage
The d.c. voltage, found by integration of the waveforms in Fig. 7.3(ii), may be
expressed in the form:

+ U,,Ccos(C,+a, -C1)-cos(C,+a, -C1)]

+ U2, Ccos (C, + a, - C,) - cos (C, + a, + n - C,)]
- UXCl + x c 2 + XC3)1
where Xci is the commutation reactance for phase i.
D.C. interconnection
An equation is derived for each converter, from the d.c. system topology relating
the d.c. voltages and currents, i.e.
f (Vd, Id) = 0 (7.3.5)
For example, the system shown in Fig. 7.1 provides the following four equations:
Vd, + Vd, + Vd3 + Vd4 - Id,.Rd = 0
Id, - Id, =0

Id, - Id3 = 0
Id, - Id, =0

The apparent redundancy in the number of d.c. variables is due to the

generality of the d.c. interconnection.

Incorporation of control strategies

Similarly to the single-phase case, any function of the variables is a valid
(mathematically) control equation so long as the equation is independent of all
the others.
Detailed consideration of the alternative firing controls is of particular interest
in this respect.
With reference to symmetrical firing control, one equation results from the
specification of minimum firing angle control, i.e.

For a six-pulse unit, the interval between firing pulses in specified as 60". This
provides two more equations.
In the equation above, phase (i) is selected during the solution procedure such
that the other two phases will have, in the unbalanced case, firing angles greater
than amin.
With conventional phase angle control, the firing angle on each phase is
specified as being equal to a,,,, i.e.
a, - amin= O (7.3.6)
a, - a,,, =0 (7.3.7)
a j - a,,, = 0 (7.3.8)
The remaining three-control equations required are derived from the operat-
ing conditions. Usually, the off-nominal taps are specified as being equal, i.e.

The final equation will normally relate to the constant current or constant
 power controller, e.g.

Inverter operation with minimum extinction angle

In contrast to the single-phase load flow, for three-phase inverter operation it is
necessary to retain the variable a in the formulation, as it is required in the
'
specification of the symmetrical firing controller. Therefore, the restriction upon
the extinction advance angle y, requires the implicit calculation of the com-
mutation angle for each phase.
Using the specification for y, as defined in Fig. 7.3 the following expression
applies :

cos yip + cos a , - I, (Xcl + XC3) = o

fiu13

Similar equations apply to the other two phases with a cyclic change of suffices.

Enlarged converter model

The three-phase equations so far developed are an exact parallel of the four
variable sequential version which was discussed in Chapter 6.
The mathematical model of the converter includes the formulation of
equations (7.2.3)and (7.2.4)for the individual phase real and reactive power flows
on the primary of the converter transformer. It is in connection with these
equations that the three-phase model deviates significantly from the single-phase
model developed in Chapter 6.
The calculation of the individual phase, real and reactive powers at the
terminal busbar requires the values of both the magnitude and angle of the
fundamental components of the individual phase currents flowing into the
converter transformer.
In the single-phase analysis, the magnitude of the fundamental current,
obtained from the Fourier analysis of the current waveshape on the transformer
secondary, was transferred across the converter transformer. This procedure is
trivial and the relevant equations were not included in the d.c. solution. The angle
of the fundamental component was calculated by simply equating the total real
power on the a.c. and d.c. sides of the converter.
A similar procedure may be applied to the three-phase analysis of the
unbalanced converter. In this case, however, the transfer of secondary currents to
the primary is no longer a trivial procedure due to the influence of the three-phase
transformer connection. In addition, the three-phase converter transformer may
influence the a.c. system operation, for example, a star-g-delta connection
provides a zero sequence path for the a.c. system.
The simplest way of accounting for such influence is to include the converter
transformer within the d.c. model.
model, i.e.

"ode = mi
The three-phase converter transformer is represented by its nodal admittance

where p indicates the primary side

and s the secondary side of the transformer
The 3 x 3 submatrices (Y,,,etc.) for the various transformer connections,
including modelling of the independent phase taps, were derived in Chapter 2.
The inclusion of the converter transformer within the d.c. model requires
twelve extra variables, as follows:
Ei& the fundamental component of the voltage waveshape at the transformer
secondary busbar;
1 the fundamental component of the secondary current waveshapes; where
i = 1,3 for the three phases.
Thus a total set of twenty-six variables is required for each converter in the d.c.
system model, fourteen of which have already been developed in previous
sections.

Remaining twelve equations

With reference to Equation (7.3.14), and on the assumption of a lossless
transformer (i.e. Y,, =jb,,,etc.) the currents at the converter side busbar are
expressed as follows:
3
=
Iiejwi - 1 [jb;: Ek&Ok+jb: v;,, ej(Bktcrm
-@term1 (7.3.15)
k= 1

By subtracting #, in the above equation, the terminal busbar angles are related
to the converter angle reference.
Separating this equation into real and imaginary components, the following six
equations result:
3
Iicosoi = 1 [bit E, sin 4, + bit V:,,
k= 1
sin (Ofer, - O:,,)] (7.3.16)

Iisin oi= 1 [ - b:
k= 1
E, cos 4, - b;t Vf,,, cos (Of,,, - O:,,)] (7.3.17)

Three further equations are derived from approximate expressions for the
fundamental r.m.s. components of the line current waveforms as shown in
Fig. 7.3, i.e.
4.1,
li= 0.995 -sin (Ti/2) (7.3.18)
Jr
where Ti is the assumed conduction period for phase i.
 The sum of the real powers on the three phases of the transformer secondary
may be equated to the total d.c. power, i.e.
3
C E,.I,-cos(q5i - a,)- Vd.Id= 0 (7.3.19)
i=l

The derivation of the last two equations is influenced by the position of the
fundamental frequency voltage reference for the secondary of the converter
transformer.
The voltage reference for the a.c. system is earth, while in d.c. transmission the
actual earth is placed on one of the converter d.c. terminals. This point is used as a
reference to define the d.c. transmission voltages and the insulation levels of the
converter transformer secondary windings.
In load-flow analysis, it is possible to use arbitrary references for each
converter unit to simplify the mathematical model. The actual voltages to earth, if
required, can then be obtained from knowledge of the particular configuration
and earthing arrangements.
With a star-connected secondary winding an obvious reference is the star point
itself. If the nodal admittance matrix is formed for a star-g-star-g connection then
this reference is implicitly present through the admittance model of the
transformer. In this case, however, the converter transformer does not restrict
the flow of zero-sequence currents and the following two equations may be
written :
3
C I,lo,=O (7.3.20)
i= 1

These two equations (real and imaginary parts) complete the set of twelve
independent equations in terms of the twelve additional variables.
However, the above considerations do not apply to delta-connected secondary
windings.
To obtain a reference which may be applied to all transformer secondary
windings, an artificial reference node is created corresponding to the position of
the zero-sequence secondary voltage. This choice of reference results in the
following two equations:
3
E , c o s ~= O~ (7.3.21)
i= 1

The nodal admittance matrix for the star-connected transformer szcondary is

now formed for an unearthed star winding. The restriction on the zero-sequence
current flowing on the secondary is therefore implicitly included in the
transformer model for both star and delta connections.
For a star-connected secondary winding both alternatives yield exactly the
same solution to the load-flow problem.
Summary of equations and variables
The twenty-six equations (R) which define the operation of each converter are:

3
R(2) = Eisin 4i= 0
i= 1

4 I d sin ( T 2 / 2 )
R(5) = I, - -.---
nfi

3
R(9) = 13.cosw 3 - 1 [b: Eksin 4,+ b,3,k.VFermsin (O:,,, - 8:e,m)]
k= 1

3
R(11) = I,.sin w 2 x 1 [b:: E kcos 4,+ b:;. v:,, cos (qerm
- O;,,)]
k= 1

I depend on transformer connection

depend on the control specifications

 R(25) = Vdan- J ~ u , ,[ C O S ( ~ , a, + - C3 + X)- c0s(C2 + U2 - C3 + n)]
-J~u,~[cos(c, +a, - c,) +
- COS(C, a, - c,)]
- J ~ u , ,[cos (c, + a, - c,)- cos (c, + a, + n - c,)]
+ Id(Xc, + Xc2 + Xc3)
R(26) =f (Vdi,Id,) from d.c. system topology.
The twenty-six variable vector (2) is:
[El, E2, E 3 , 4 1 , 4 2 , 4 3 , 1 1 , 1 2 , 1 3 , 0 1 , 0 2 , 0 3 ,
u ~ ~ , u C1,C2,C3,a1,a2,a3,al,a2,a3,
~ ~ , u ~ ~ ,
vd, 'dl

7.4. LOAD-FLOW SOLUTION

A sequential technique, using the t hree-phase fast-decoupled a.c. algorithm and a
full Newton-Raphson algorithm for the d.c. equations, involves the block
successive iteration of the following three equations,

where [B'] and [ B ] are the three-phase fast-decoupled a.c. Jacobian matrices as
developed in Chapter 5, and [J] is the d.c. Jacobian of first order partial
derivatives.
Equations (7.4.1) and (7.4.2) are the three-phase fast-decoupled algorithmic
equations from Chapter 5. For the solution of the equations (7.4.1) and (7.4.2), the
d.c. variables 2 are treated as constants and, in effect, the d.c. system is modelled
simply by the appropriate real and reactive power injections at the converter
terminal busbar.
These power injections are calculated from the latest solution of the d.c. system
equations and are used to form the corresponding real and reactive power
mismatches. For the d.c. iteration, the a.c. variables at the terminal busbars are
considered to be constant.
The iteration sequence for the solution of equations (7.4.1), (7.4.2) and (7.4.3) is
illustrated in Fig. 7.5. It is based on the P, Q, DC sequence described in Chapter 6
which proved the most successful sequential technique in the single-phase case.
This sequence acknowledges the fact that the converter operation is strongly
related to the magnitude of the terminal voltages and more weakly dependent on
their phase angles. Therefore, the converter solution follows the update of the a.c.
 1Evaluate real power mismatches]

Solve equatiy (74.1)

and update 6

voltage regulator mismatches

1
"' Evaluate d.c. residuals

I
t
Solve equation ( 7.4.3 )
and update i
1 I
kP=O

Fig. 7.5. Flow chart for threephase a.c.-d.c. load flow

terminal voltages. It should be noted, however, that in the three-phase case, final
convergence is comparatively slow because the d.c. system behaviour is
dependent on the phase-angle unbalance as much as on the voltage unbalance.

7.5. PROGRAM STRUCTURE AND COMPUTATIONAL ASPECTS

The main compbnents of the computer program are illustrated in Fig. 7.6. The
additional blocks and increase in size of the a.c.-d.c. program over the purely a.c.
 A x . data input
Form and store I

-initialize d.c.
variables (410)

Iterative solution
procedure. (Fig. 7.5)
(1325)

Fig. 7.6.

algorithm may be assessed by comparison with Fig. 5.4. The numbers in

parenthesis are the approximate number of FORTRAN statements. The
additional features are discussed in the following sections.

D.C. input data

The input data for the d.c. system consists of the parameters of each converter
including maximum and minimum variable limits where appropriate. In
addition, the d.c. network equations (7.3.5) must be formed from the d.c. system
topology. As the d.c. system is relatively small and simple in its interconnection
these equations are formed by inspection and effectively input directly by the user.
The d.c. system variables jl are initialized as the balanced three-phase
equivalent of the single-phase converter variables as discussed in Section 6.8.

Programming aspects of the iterative solution

The iterative solution (Fig. 7.5) for the a.c.-d.c. load flow is significantly enlarged
over the purely a.c. case (Section 5.5). The basic reason is that the d.c. Jacobian
must be reformed and refactorized at each iteration. In addition, because of the
nonuniform nature of the d.c. Jacobians and residual equations, each term must
be formulated separately in contrast to the a.c. case where compact program
loops may be used.
Equations (7.4.1) and (7.4.2) are solved using sparsity techniques and near
optimal ordering as described in Chapter 4. Similarly to the single-phase case, the
equations for each converter are separate except for those relating to the d.c.
interconnection and the solution of equation (7.4.3) is carried out using a
modified Gaussian Elimination routine.
This feature may be utilized by appropriate ordering of variables to yield a
block sparsity structure for. the d.c. Jacobian. With this aim, the d.c. voltage
variable is placed last for each block of converter equations and all the d.c.
current variables are placed after all converter blocks. The d.c. Jacobian will then
have a structure as illustrated in Fig. 7.7.

Fig. 7.7. Jacobian structures for four converter d.c. system

(nonzero elements indicated)

By using row pivoting only during the solution procedure, the block sparsity of
Fig. 7.7 is preserved. Each block containing nonzero elements is stored in full, but
only nonzero elements are processed.
This routine requires less storage than a normal sparsity program for
nonsymmetrical matrices and the solution efficiency is improved.

7.6. PERFORMANCE OF THE ALGORITHM

Test system
The performance of the algorithm is discussed with reference to the test system
illustrated in Fig. 7.8. The system consists of two a.c. systems interconnected by a
600 kV, 600 MW h.v.d.c. link.
The twenty-bus system is a representation of the 220 kVa.c. network of the
South Island of New Zealand. It includes mutually-coupled parallel lines,
 Twenty-bus system Five - bu; system
(a)

BUS 01
I

BUS 05

Fig. 7.8. Three-phase a.c.-d.c. test system. (a) H.V.D.C. interconnection;(b) Five-bus a.c.
system

synchronous generators and condensers, star-star and star-delta connected

transformers and has a total generation in excess of 2000 MW.
At the other end of the link, a fictitious five-bus system represents 800 MW of
remote hydrogeneration connected to a converter terminal and load busbar by
long untransposed high voltage lines.
The small system is used to test the algorithm and to enable detailed discussion
of results. The d.c. link should have considerable influence, as the link power
rating is comparable to the total capacity of the small system. Relevant
parameters for the a.c. system and d.c. link are given in Table 7.1.

Convergence of d.c. model from fixed terminal conditions

Typical convergence patterns for the terminal power flows for the three-phase
model, under both balanced and unbalanced terminal conditions, are shown in
Fig. 7.9. The convergence pattern of the single-phase algorithm developed in
 Table 7.1. System data
Data for all lines. Data for generator transformers

Z, Series impedance matrix Connection Star-G/Delta

Reactance +
0.0016 j0.015
0.0066 0.0017 0.0012 Off-nominal tap + 2.5% on star
+ jO.056 +jO.027 +j0.021
0.00 17 0.0045 0.00 14
+jO.027 +jO.047 +j0.022
0.0012 0.0014 0.0062
+ ~0.021 +j0.0220 +j0.061

Y, shunt admittance matrix

Data for all converters

Phase 1 Phase 2 Phase 3

Transformer reactances 0.0510 0.0510 0.0510

Commutation reactances 0.0537 0.0537 0.0537
Minimum firing angle 7.0 deg
Minimum extinction angle 10.0 deg
Nominal voltage 140 kV

D.C. link resistance = 25.0 ohms.

Generator data

Sequence
reactances Voltage
Power regulator
Name Xo x I *, (MW) Va

GENOl 0.02 - 0.004 700.0 1.045

GENSL 0.02 - 0.004 Slack 1.061

Busbar loadings

Phase A Phase B Phase C

Pus name P-load Q-load P-load Q-load P-load Q-load
 - 1 2 3 4 lteration ( a ] I 2 3 4 Iteration

1 2 3 4lteration(b) 1 2 3 4 Iteration

Fig.7.9. Convergence of terminal powers for three-phase converter model.

(a) Unbalanced; (b) Balanced

Chapter 6, is also illustrated. To enable a comparison to be made, the total three-

phase powers are plotted for the balanced case. In all cases the d.c. starting values
were selected to give large initial errors in the terminal powers to better illustrate
the convergence pattern.
The d.c. equations require two iterations to converge for both the single and
three phase models.

Performance of the integrated a.c.-d.c. load flow

With reference to the test system illustrated in Fig. 7.8, the following control
specifications are used at the inverting terminal for all test cases
- Symmetrical firing control with the reference phase on minimum extinction
angle.
- Off-nominal tap ratios equal on all phases.
- D.C. voltage specified.

A variety of different control strategies are considered at the rectifier terminal

and the convergence results are given in Table 7.2. For comparison, the table
includes cases with the converters modelled as equivalent a.c. loads.
It should be noted that the iteration scheme illustrated in Fig. 7.5 does not
allow for each individual a.c. system to be converged independently, therefore, the
number of iterations required is the larger of the two sets given in the table.
It is clear that the integration of the d.c. converter model does not cause any
significant deterioration in performance. The only cases where convergence is
 Table 7.2. Case descriptions and convergence results
Number of iterations to
case Case description and rectifier specifications Convergence (0.1 M W I M V A R )
--
20-bus system 5-bus system

49 Converter modelled by equivalent balanced loads* &7 6,s

(ii) Converter modelled by equivalent unbalanced loads* 8,7 6,s
Hi) Phase-angle control; a , = a , = a, = a,,,, a , = a, = a,, Pdc= Pip
(ii) Symmetrical firing; ai = a,,
(iii) a , = a, = a,, I , = I::,
Phase-angle control; a , = a, = a , = amin, vdl = Vd2

(iv) Symmetrical firing; ai = a,, 8,7 6,s

(4 As for case HI); with poor starting values. (Pd,,Q, in error by 70%) 8,7 8,7

(vi) As for case Mi); with large unbalanced load at BUS03 8,7 7,6
(vii) As for case b(ii); with large unbalanced load at BUS03
(viii) As for case Hi); with loss of 1 line BUS01 to BUS03
(ix) Symmetrical firing; ai = a,,,,a, = - 10%,a, = O,a, = + 10% 7,6
(XI Phase-angle control; a , =a, =a, = d p ,a , = a, = a,, p,, = P z 8,7 7,6

(4 Case (x) loss of 1 line. BUS01 to BUS03 8,7 88

Loading for case a(i) and a($ derived from results for case b(i). See Table 7.3.
 I I
1
I
2
I
3
I
4
I
5
I
6
--
Iteration
(a1

Fig. 7.10. Convergence patterns of terminal conditions for a

strong ax. system (a) A.c. terminal voltages; (b) Terminal
reactive power flows; (c) A.c. terminal angle unbalance (de-
viation from nominal)

slowed are (viii) and (xi) where the system is weakened by the loss of one
transmission line. This is to be expected from the discussion of single-phase
sequential algorithms given in Section 6.8.
To examine the effect of a weak system in the three-phase case, the convergence
patterns for the terminal powers and voltages are shown for case (xi) in Fig. 7.10.
The reactive power and voltage unbalance vary considerably over the first few
iterations but this initial variation does not cause any convergence problems.
With weaker systems, the unbalance increases and the convergence patterns
become more oscillatory. The corresponding convergence pattern of the single-
phase load flow for case (xi) is shown in Fig. 7.1 l(b) where a similar oscillatory
pattern is observable. Moreover the sum of the three-phase reactive powers and
 x-X-X-X %rm
I (tve sequence)

Fig. 7.1 1. Comparison of single-phase and three-phase po-

sitive sequence convergence patterns. (a) Three-phase load
flow; (b) Single-phase load flow

the average phase voltage of Fig. 7.10, plotted in Fig. 7.1 l(a), shows an even
closer similarity between the three-phase case and the single-phase behaviour.

Sample results
.The operating states of the two converters connected to BUS03 are listed for the
most typical cases in Table 7.3. The corresponding a.c. system voltage profiles
and generation results are given for cases a(i), b(i) and b(ii) in Table 7.4. The
following discussion is with reference to these results.
The results of the realistic three-phase converter model (case b(i)), although
 Table 7.3(a) Converter 1 results

Converter 1 (Star-Star)

Commu- Terminal powers Dc. conditions

Firing Tap tation
angle ratio angle Real Reactive Voltage Current
Case Phase ai (deg) ai(%) u, (deg) pi (MW) Q, (MVAr) Vd, (kv) I d , (kA)

Table 7 4 b ) Converter 2 results

Converter 2 (Star-G-Delta)

Terminal powers Dc. conditions

Commu-
Firing tation
angle angle Real Reactive Voltage Current
Case Phase a, (deg) ui (deg) Pi (MW) Qi (MVAr) V d , (kV) I d ,
7.00 29.80
7.00 29.60
7.00 29.32
8.03 28.97
7.00 29.57
8.55 28.08
7.00 30.63
7.00 28.92
7.00 28.90
7.00 30.48
14.95 23.25
13.41 24.25
8.08 25.42
8.38 27.30
7.00 26.96
 Table 7.4. Bus voltages and generation results
Case a(i)

Phase A Phase B Phase C Generation

Bus name Volt. Ang. Volt. Ang. Volt. Ang. Total

BUS01 1.067 27.294 1.067 -92.891 1.061 147.431 0.000 0.000

BUS02 1.054 25.190 1.065 - 94.670 1.057 144.915 0.000 0.000
BUS03 1.038 23.185 1.071 - 95.714 1.043 142.567 0.000 0.000
BUS04 1.045 - 3.566 1.046 - 123.479 1.047 116.436 173.621 74.723
BUS05 1.061 2.683 1.062 -117.367 1.061 122.628 700.000 113.920

Case Mi)

Phase A Phase B Phase C Generation

Bus name Volt. Ang. Volt. Ang. Volt. Ang. Total

BUS01 1.067 27.362 1.065 - 92.955 1.062 147.437 0.000 0.000

BUS02 1.055 25.232 1.064 -94.717 1.057 144.925 0.000 0.000
BUS03 1.038 23.517 1.066 - 95.965 1.049 142.543 0.000 0.000
BUS04 1.045 - 3.552 1.046 - 123.483 1.047 116.438 173.570 74.706
BUS05 1.061 2.690 1.062 - 117.369 1.060 122.634 700.000 113.680

Case Mii)

Phase A Phase B Phase C Generation

Bus name Volt. Ang. Volt. Ang. Volt. Ang. Total

BUS01 1.066 27.31 1.066 - 92.942 1.062 147.421 0.000 0.000

BUS02 1.054 25.238 1.064 - 94.705 1.057 144.913 0.000 0.000
BUS03 1.036 23.532 1.066 - 95.947 1.049 142.506 0.000 0.000
BUS04 1.045 - 3.563 1.046 - 123.479 1.047 116.439 173.593 75.949
BUS05 1.061 2.690 1.062 - 117.363 1.060 122.635 700.000 115.391

distinguishable from those of the balanced model a(i), are not significantly
different as regards the a.c. system operation. They are definitely significant
however, as regards converter operation, particularly when consideration is
given to the harmonic content.
A comparison of cases b(i) and b(ii) shows an increase in reactive power
consumption in case b(ii) due to two phases having greater than minimum firing
angles.
The results also show that the transformer connection modifies the converter
source voltages and the phase distribution of power flows. Under balanced
conditions, a zero-sequence voltage may appear at system busbars. As the
converter has no zero-sequence path, zero sequence current will only flow when
the converter transformer provides a path, as in the case of the star-G/delta
transformer. A typical example is illustrated in Fig. 7.12 where the zero-sequence
voltages and currents are shown for case b(i). Accurate converter transformer
models must therefore be included in the converter modelling.

Conclusions on performance of the algorithm

The fastdecoupled three-phase a.c.-d.c. load flow behaves in a very similar
 I b
'0
- m
where = jO.051
lab

Fig. 7.12. Sequence components and the converter transformer con-

nection. (a) Zero sequence potentials for case b (i); (b) Zero sequence
network for converter transformers. (Note: Transformer secondary
zero sequence reference is provided by equations (7.3.21)and (7.3.22))

manner to the corresponding single-phase version. The following ,general

conclusions can be made on its performance:
- the number of iterations to convergence is not significantly increased by the
presence of the d.c. converters;
- d.c. convergence is not dependent on the specific control specifications
applied to each converter;
- wide errors in initial conditions may be tolerated;
- for very weak a.c. systems the interaction of the converter with the a.c. system
is increased and the convergence is slowed. Successful convergence can,
however, be expected in all practical cases;
- the algorithm exibits good reliability even under extreme unbalance.

7.7. EFFECT OF HARMONIC VOLTAGES ON THE FUNDAMENTAL

FREQUENCY OPERATING STATE OF D.C. CONVERTERS
The three-phase model of the d.c. converter has been formulated under the
assumption of perfectly sinusoidal voltages at the converter terminal busbar.
Without this assumption the steady-state load-flow formulation would be very
complex due to the difficulty of calculating accurately the level of harmonic
voltages which may be present.
However, the presence of harmonic voltages at the converter terminal will alter
the fundamental frequency power flows primarily because the actual zero
crossings of the phase-to-phase voltages will be shifted from those calculated
from the fundamentals alone. Other effects (e.g. modification of average d.c.
voltages and commutation time) are of secondary importance and may be
ignored.
The effect of harmonic voltages is also influenced by the control system in
operation. However, the order of magnitude of any possible errors may be
assessed by assuming that the control system, whether phase-angle control or
symmetrical firing, is perfect (i.e. there is no firing angle error) and is presented
with the actual zero crossings of the commutating voltages (i.e. there are no
control system filters).

Harmonic voltage limits

In order to reduce the undesirable effects caused by the harmonic voltages and
currents the supply authorities specify maximum permissible values for the
harmonics.
Typical recommended limits for harmonic voltages in percentage of the
fundamental are as follows:
Supply voltage Odd Even
415 V 4 2
33kV 2 1
110/132kV 1 0.5
In addition to limits on individual harmonics the maximum total harmonic
distortion (THD) is also specified.
The total harmonic distortion is defined by:

and a typical limit for a 110/132-kV network is 1.5 percent.

Shift in zero crossing of phase-phase voltages

There is considerable difficulty in assigning a realistic worst case for the
investigation of the influence of harmonic voltages. The following considerations
apply:
- Under balanced conditions the characteristic harmonic orders do not
influence the intervals between firing pulses, i.e. all firings are shifted by equal
amounts. Therefore these voltages will have very little effect on the magnitude
of the fundamental currents.
- Harmonic limits are usually applied at the point of common coupling to the
 supply network which may or may not be the converter terminal busbar.
- The majority of the allowed triplen harmonics will consist of the usual zero
sequence components and these have no influence on the position of phase to
phase crossings.
The order of the nonzero sequence triplen harmonics, which may be considered
to cause the most significant shifts, is not important as, under balanced
conditions with the subsequent crossings 120 degrees apart, a small magnitude of
any lower order triplen harmonic will result in comparable shifts for all three
phases.
As a consequence of these features, it is considered reasonable to investigate the
worst case effect of 2 percent of nonzero sequence third harmonic voltage.
The worst case for a shift in any zero crossing is for all harmonics to add to the
fundamental voltage on one phase and to subtract from the other phase at the
position of actual zero crossing.
With consequent zero crossings 60 degrees apart, i.e. one-half rotation of the
third harmonic phasors, then one subsequent shift will be approximately half the
initial worst case, and the third shift will be equal to the worst case initial shift
except in the opposite direction.
For 2 percent nonzero sequence third harmonic, the worst case shift is
approximately one degree.
The order of magnitude of the resultant errors can be assessed on the basis of
nominally balanced operation and by interpreting shifts in zero crossings as firing
angle errors determined in accordance with the firing controller in operation.
For a converter operating with fixed tap ratios and specified d.c. current, any
firing-angle errors are reflected into the converter operation through errors in the
calculated d.c. voltage and in the calculated magnitudes for the fundamental
component of the phase current waveforms. Both errors occur to some extent
with both firing controllers.

Effect of crossing error on the average d.c. voltage

The largest error occurs with symmetrical firing control where any shift in the
reference is reflected as a change in firing angle on all phases. The d.c. voltage
error depends upon the nominal firing angle. Typical errors in d.c. voltage under
symmetrical firing control are illustrated in Table 7.5 on the assumption of
balanced fundamental voltages. Although the errors are small at normal firing
angles, they are significant at large firing angles.
Table 7.5. Percentage error in calculated d.c. voltage (symmetrical firing)

a
(degrees) Firing angle errors due to shift of zero crossings
Errors in calculated phase current magnitudes
Appreciable errors occur with the case of phase-angle control; errors with
symmetrical firing are limited to the effects of commutation angle unbalance.
Theeffect of up to two degrees modulation has been investigated on the basis of
the current waveform of a converter with commutation angle of ten degrees and
balanced voltages. Ignoring the shifts in commutation angle which will occur
with an alteration of firing angle, the effect of alteration in the period of
conduction has been investigated using a Fourier Transform algorithm. The
results are shown in Table 7.6. In addition to the fundamental, the percentages of

Table 7.6. Harmonic currents

120" 121" 122" 120" 121" 122"

the harmonics are also given. Note that no even harmonics are present as the
waveform was assumed to be symmetrical.
It is important to note that the percentage errors presented in the Table are the
maximum that can occur; in all practical cases there will be an alteration in the
commutation angle which will inhibit the change in the waveform. This effect will
be most noticeable at small firing angles. In general therefore, the errors will be
less than those indicated in the table.
The change in fundamental is not, therefore, expected to exceed 1 percent of the
value calculated by the load flow.

7.8. REFERENCES
1. J. D. Ainsworth, 1967. 'Harmonic instability between controlled static converters and
a.c. networks', Proc. IEE, 114 (7), 949-957.
2. J. D. Ainsworth, 1968. 'The phase-locked oscillator-A new control system for
controlled static converters', I EE E, PAS, PAS-87 (3), 859-865.
3. J. Kauferle, R. Mey and Y. Rogowsky, 1970. 'H.V.D.C. stations connected to weak a.c.
systems', I EEE, PAS, PAS49 (7), 1610- 1617.
4. A. G. Phadke and J. H. Harlow, 1966. 'Unbalanced Converter Operation', IEEE
PAS, P A S S , 233-239.
5. J. Arrillaga and A. E. Efthymiadis, 1968. 'Simulation of converter performance under
unbalanced conditions', Proc. IE E, 115 (12), 1809-1 818.
 Faulted System Studies
8.1. INTRODUCTION
The main object of Fault Analysis is the calculation of fault currents and voltages
for the determination of circuit-breaker capacity and protective relays
performance.
Early methods used in the calculation of fault levels involved the following
approximations:
- All voltage sources assumed a one per unit magnitude and zero relative phase,
which is equivalent to neglecting the prefault load current contribution.
- Transmission plant components included only inductive parameters.
- Transmission line shunt capacitance and transformer magnetizing imped-
ance were ignored.
Based on the above assumptions, simple equivalent sequence impedance
networks were calculated and these were interconnected according to the fault
specification. Conventional circuit analysis was then used to calculate the
sequence voltage and currents and with them, by means of the inverse sequence
component transformation, the phase components.
Although the basic procedure of the computer solution is still the same, the
need for the various approximations has disappeared.
The three-phase models of transmission plant developed in Chapter 2, which
included interphase and parallel line mutual effects, could be easily combined to
produce the faulted system matrix admittance or matrix impedance and hence
provide an accurate model for the analysis of a.c. system faults.
However, the main reasons given for the use of the phase frame of reference in
load flows are less relevant here. Extra losses and harmonic content are less of a
problem in the short period of time prior to fault clearance. Fault studies are
normally performed on systems reasonably well balanced either at the oper-
ational or planning stage; in the latter case only after prospective system
configurations have been proved acceptable through load-flow studies.
Moreover, unbalanced faults involving large converter plant, while in need of a
three-phase representation, cannot be assessed by steady-state models as
explained in Chapter 11.
 Finally, faulted system studies constitute an integral part of multi-machine
transient stability programs, the complexity of which will not normally permit
the three-phase approach.
A single-phase representation, achieved with the help of the symmetrical
components transformation(') is used in this chapter as a basis for the
development of a fault-study p r ~ g r a m . ( ~ ) . ( ~ ) . ( ~ ) ~ ( ~ )

8.2. ANALYSIS OF THREE-PHASE FAULTS

A preliminary stage to the analysis is the collection of appropriate data specifying
the system to be analysed in terms of prefault voltage, loading and generating
conditions. Such data is then processed to form a nodal equivalent network
constituted by admittances and injected currents.
The equivalent circuits of loads, lines and transformers discussed in Chapter 2
are directly applicable here. The generators can be represented by a constant
voltage E~ behind an approximate machine admittance y M which value depends
on the time of the calculation from the instant of fault inception. This is illustrated
in Fig. 8.l(a).

Fig. 8.1. Generator representation

When analysing the first two or three cycles following the fault, the
subtransient admittance of the machine is normally used, whilst for longer times,
it is more appropriate to use the transient admittance. The machine model,
illustrated in Fig. 8.l(a), is then converted to a nodal equivalent by means of
Norton's Theorem which changes the voltage source into a current source
injected at the bus j as shown in Fig. 8.l(b). This is most effective as otherwise a
further node at j' is necessary to define the machine admittance yM.
The injected nodal current is given by

where

so that
 17 is the current required at the voltage V, to produce the machine power
P7 +j Q 7 , so
(I?)* Vj = P y +j - Q r (8.2.4)
Thus from the load-flow data of P M ,Q~ and V Mwe may calculate the injected
nodal current I ; , as

Admittance matrix equation

Let us take as a reference the small system of Fig. 8.2. Each element is converted to
its nodal equivalent. These are connected together as shown in Fig. 8.3 and finally
simplified to the equivalent circuit of Fig. 8.4.

Fig. 8.2. Example of small power system

The following equations may then be written for the network of Fig. 8.4.

or in matrix form after grouping together the terms common to each voltage:
Fig. 8.3. Model substitution

Fig. 8.4. Final equivalent

184
where

Equation (8.2.1 1) is usually written as

where [a and [V] are the current and voltage vectors and [Y] is the nodal
admittance matrix of the system of Fig. 8.2.
It can be seen from equations (8.2.6) to (8.2.10) that nonzero elements only
occur where branches exist between nodes. Since each node or busbar is normally
connected to less than four other nodes, there are usually quite a number of zero
elements in any system with more than ten busbars. Such sparsity is exploited by
only storing and processing the nonzero elements. Moreover, the symmetry of the
matrix (Yij = Yji) permits using only the upper right hand terms in the
calculations.

Impedance matrix equation

The nodal admittance equation is inefficient as it requires a complete iterative
solution for each fault type and location. Instead, equation (8.2.12) can be written
as

[v] = [Y]-~.[I]
=P I . [I1
This equation uses the bus nodal impedance matrix [Z] and permits using the
Thevenin Equivalent circuit as illustrated in Fig. 8.5, which, as will be shown
later, provides a direct solution of the fault conditions at any node. However, the
use of conventional matrix inversion techniques results in an impedance matrix
with nonzero terms in every position Zij.
The sparsity of the [Y] matrix may be retained by using an efficient inversion
and the nodal impedance matrix can then be calculated directly
from the factorized admittance matrix.

Fault calculations
From the initial machine data, the values of [aare first calculated from equation
(8.2.5) using one per unit voltages. These may now be used to obtain a better
estimate of [V], the prefault voltage at every node from equation (8.2.13). If the
initial data is supplied from a load flow, this calculation will not make any
difference.
The program now has sufficient information to calculate the voltages and
currents during a fault.
 The voltage at the fault bus k is from Fig. 8.6.
I/{ = z * . I*
where
k is the bus to be faulted,
Z* is the fault impedance and
I* is the fault current.

Fig. 8.5. Thevenin equivalent of prefault system

Fig. 8.6. Thevenin equivalent of faulted system

Equation (8.2.13) may be expanded to yield

Selecting row k and expanding:

'k =zklll + Zk2
This equation describes the voltage at bus k prior to the fault. During a fault a
large fault current I* flows out of bus k. Including this current in equation (8.2.16)
and using equation (8.2.14)
and so the fault current is given directly by

Also from equation (8.2.15)the prefault voltage at any other bus j is

and during the fault

From equations (8.2.19) and (8.2.22) the fault voltages at every bus in the
system may be calculated, each calculation requiring only one column of the
impedance matrix. Using the bifactorization routines the kth column can be
obtained by multiplying the impedance matrix by a vector which has a '1' in the
kth row and '0"s elsewhere, i.e.

Once Z,, is known then I f is calculated from equation (8.2.19). I f is then

subtracted from the initial prefault nodal currents to form a new vector [I*]
defined by
I J/ = I J. for j f k, j = 1 to n
I{ = I, - If for j = k
The voltages during the fault are given by the product of the impedance matrix
and this new vector [ I f ] , i.e.
[v*] = [ Z ] . [ I f ] (8.2.24)
Equation (8.2.24)is equivalent to (8.2.22)because of the following expansion.
[ r f ] = [ q - [ o , o , o ,... r f ,... olT
from which equation (8.2.24)expands as
[v*] = [ Z ] { [ I ] - [O,O,O,... If ,... 0 l T )
 [Vf] = [V] - [Z]-[O,O,. . .If,. . .OIT
which is equivalent to equation (8.2.22).
Once the fault voltages are known the branch currents between buses can be
calculated from the original branch admittances, i.e.

A correction is necessary for the sending end current of a tapped transformer,

i.e.
IL = yij{(l - T)V{ - V;) (8.2.26)
With reference to Fig. 8.7, a machine fault current contribution is
1M.f = (EM - I/{)- yM
or substituting Ii = y?. EM (from equation (8.2.1) ).

8.3. ANALYSIS OF UNBALANCED FAULTS

If the network is unsymmetrically faulted or loaded, neither the phase currents
nor the phase voltage will possess three-phase symmetry. The analysis can no
longer be limited to one phase and the admittance ofeach element will consist of a
3 x 3 matrix which on the assumption of a reasonably balanced tranmission
system, will be symmetrical, i.e.

Fig. 8.7. Machine representation showing fault current

contribution
Matrix (8.3.1) can be diagonalized by the symmetrical components transfor-
mation (T*)'Y T into its sequence component equivalent, i.e.

where

Moreover, for stationary balanced system elements the admittances Ob Y and

"'Y are equal and equations (8.3.3) show that the corresponding positive and
negative sequence admittances are also equal. Further, the simplifying assum-
ption is often made that the positive and negative sequence admittances of
rotating machines are equal. This assumption is only reasonable when the
subtransient admittances are being used and in such case the storage required by
the program can be substantially reduced by deleting the negative sequence
matrices.

Admittance matrices
The data specifying each element of the system is then used to form the following
three nodal equations:
OIi = O V i O y i i + ( O V i + O V l ) O y .i .i .++ ( O V i - O V n ) O ~ n i (8.3.4)
' I i = ' V i ' ~ i i + ( ' V i - ' V l ) ' ~ i i -+. . + ( ' V i - ' V n ) ' ~ n i (8.3.5)
2 1 i = 2 ~ i 2 y i i + ( 2 V 2i -~ l ) 2 y i i + . , . + ( 2 V 2Vn)2yni
i- (8.3.6)
where
' I i is the zero-sequence injected current at bus i.
' V i is the positive-sequence voltage at bus i.
and
2yniis the negative-sequence admittance between nodes n and i.
The above equations can be expressed as:
[OI] = [O Y ] [OV]
[ ' I ] = ['Y ] [' V ]
C21] = C2 Y ] C2 V ]
 where

and
n
YYii = C Yyik for i = 1, n; y = 0,l, or 2
k= 1

The sequence-admittance matrices can now be triangularized by the bifactori-

zation method. Since the three admittance matrices have identical structure, this
can be made more efficient by triangularizing them simultaneously, i.e. in
programming terms, only one set of vectors is needed to form pointers to the three
arrays as they are stored by the bifactorization routine.

Fault calculations
As already explained for the three-phase fault, the nodal impedance matrices may
now be calculated directly from the reduced admittance matrices and the
following sequence impedance matrix equations result.

[O V] = [OZ] [Ol]

['V] = ['z] ['I]

C2 V] = C2Z] [21]

Because the system is assumed to be balanced prior to the fault, the vectors of
negative and zero-sequencecurrents are zero, i.e. there are no prefault negative or
zero-sequence voltages.
The positive-sequence network then models the prefault network condition
and equation (8.3.11) is used to calculate the prefault voltages. If the original
voltages used in the machine models were obtained from a load-flow calculation,
then the use of equation (8.3.1 1)will make no difference to those results; however,
if the voltages were assumed at one p.u. with zero angle then this calculation will
provide more accurate prefault voltages.
The single-phaseequivalent circuit is then set up by linking the three sequence
networks together according to the type of fault to be analysed.(*)

Short-circuit faults
A convenient way of simulating the fault location F for the analysis of short-
circuit faults is illustrated in Fig. 8.8.
It includes three fault impedances a Z , b and~ 'Z and three injected currents
'I*, bl*, and 'I*
For each type of fault, it is possible to write 'boundary conditions' for the
currents and voltages at the fault location. For example, Fig. 8.9 shows the case
of a line to ground fault at bus k.
 0
A
-
a

b .r
C
A
4
Ov bv cv

-
I
Fig. 8.8. The fault location

0
A -
b v
a
A
C a

A
c'., olf'r b;fJf; 0
"% *'.,

-
1
-
Fig. 8.9. Single line to ground fault

The boundary conditions are

blf =C l f =0

Using equations (8.3.13)and (8.3.14) with the sequence components transfor-

mation the following relationships result.

Also, the sequence voltages at the fault location may be described by the
equations
Ovf -
k -
-0 ~ ~ ~ ' O l f
(8.3.17)
lv,f =lvk- lzkk.llf (8.3.18)
2 v f= - 2 z k k . 2 1 f (8.3.19)
From equations (8.3.15) to (8.3.19), the following relationships are obtained.

Similar considerations yield the fault currents for other types of short-circuit
fault. The results for line-to-ground, line-to-line, line-to-line-to-ground, and line-
to-line-to-line faults are illustrated in Table 8.1.
Table 8.1. Fault currents for short-circuit faults

Fault 'If l1f

L-G 'If

- '1f

K - OZfi.' I f - 2Zfi.'If
G L-G
+
( 2 z ; i . 0 z ; ) / ( 2 z ; i OZ;) + 'ZIi 2Zii + OZIi 2zIi + OZIi
L-L-GG

where
+
'Zli = ' Z i i 0.5 Z*
'Z,',= ' Z i i + 0.5 Z*
OZIi = OZii + 0.5 Z f
These fault currents at the fault location are then added to the current vectors
['I], ['I] and ['I] to produce the fault current vectors [OI*], [ l I*] and ['I*]. For
a fault at bus k these are
0 for i # k
-'I* for i = k

'1-I =
{::: for i # k
- lI* for i = k
0 for i # k
- 'I* for i = k

The fault voltages are then obtained from equations (8.3.10) to (8.3.12) by
substituting the fault current vector for the prefault current vector, i.e.
[O V*] = [OZ] (8.3.24)
[l V * ] = [ l Z ] [ l l * ] (8.3.25)
[' V * ] = ['Z] ['I*] (8.3.26)
Open-circuit faults
The system is now represented by a two-port network across which the faulty line
is connected as shown in Fig. 8.10. In this case, the prefault voltages have to be
obtained from a load-flow study.

System

Oz * 01'
q, * dl'
A
A
A
Cz CIf
A
+ A
ck$ *k$ '5 dl(

-
-L

Fig. 8.10. Two-part network with faulty line where " Z ,bZ
or 'Z may be on open circuit

For an open-circuit fault on phases '6' and 'c' the boundary conditions are

Using these equations with the sequence transformation, the following

relationships result
OI/ = 'I/ = ZI/ = 01/13 (8.3.27)

Equations (8.3.27) and (8.3.28) define the connection of the Thevenin equiva-
lent sequence networks at the fault location to solve for the fault currents.
The equivalent Thevenin impedances are the sequence impedances of the
system between the two buses k and 1, i.e.

and the equivalent Thevenin voltage is given by the difference between the
voltage at buses 1 and k with the faulted line disconnected.
During the fault the sequence voltages OV,/,,'V{~and 'V,/, have the same
expressions as equations (8.3.17) (8.3.18) (8.3.19).Thus similar considerations, as
in the case of the line-to-ground short-circuit, lead to the following expression for
the fault currents.
 The case of a single open-circuit fault can be analysed in a similar manner and
the final relevant equations are shown in Table 8.2.
The fault current vector is formed as follows:
0 for i = l , n i#korl

i=l
'Ii for i = l,n if korl
'Ik - f i =k
I ,I i= 1

and the voltage vector is given by equations (8.3.24) to (8.3.26).

Table 8.2. Fault currents for open-circuit faults

Fault IJ 21J OIJ

ONEO-C

TWOO-C

where
'z = 'zkk+ 'zll- 'zk l - l Z l k
Oz = Oz,, + Oz,,- Oz,, - Oz,,
2~ = 'Zkk + 'Zll - 2zkl- 'Zlk
zf is the sum of the positive, negative and zero sequence impedances of the
faulty circuit, i.e.
zf = 'zf + 2zf + Ozf
'Z' = 'z + 'zf
OZ' = oz + Ozf

2 ~ =' 2z + 2zf
From the fault voltages the branch currents are obtained as follows:
O I f . = OY. .(O v{ - O V
)!
11 IJ (8.3.31)
'16 = yij(l V{ - v!) (8.3.32)
21; = zYij(2I/{ - vf) (8.3.33)
 Where necessary the corrections for taps on the positive and negative sequence
networks are

Finally, the machine contributions may be calculated, i.e.

8.4. PROGRAM DESCRIPTION AND TYPICAL SOLUTIONS

A fault analysis program must be capable of analysing the following a.c. system
faults.
(1) line-to-ground short circuit;
(2) line-to-line short circuit;
(3) line-to-line-to-ground short circuit;
(4) line-to-line-to-line to ground short circuit;
(5) single open-circuit line;
(6) double open-circuit line.
However, the conventional a.c. program cannot in general be used for the
solution of similar faults in the presence of h.v.d.c. links. An alternative program
for this purpose is described in Chapter 11.
Basic to the fault study program is the determination of the impedance matrix
of the system, the elements of which can be used, along with the conditions
imposed by the type of fault, to directly solve for the fault currents and voltages.
The main steps of a general-purpose fault program are indicated in Fig. 8.11.
Prefault information and typical outputs for balanced and unbalanced fault
conditions are illustrated in the following computer printouts. The printouts
relate to studies carried out for the New Zealand South Island a.c. system.
1 Define fault type codes 1
-
-7

Read and pr~ntt ~ t l ecards

v

Read busbar data and

calculate bus admittances

Read branch data and calculate

branch admittances and bus
shunt admittances for lines and
transformers

I Generate the factorized

impedance matrix I

print options for each fault

I
II Calculate the fault currents
and print fault MVA

I Calculate the fault voltages 1

Optional calculation of branch

( Increment fault busbar number I

f

A Next fault

Fig. 8.1 1 . General flow diagram

00
00

..
UIN
oa
*N

00

N9
me

..
d
00

-..
00

09
on
Cul
0-

00

N9

..
09
rO
00
00
00

00
00

..
00
0-
00

00

2s
..
-a
00
00

00

3"i
em
'2
..
00

%
 W
O
a
J
.
C w
..................................
00000
00000
*0000
00000
0a000
00000
00000
00000
00000
00000
OOOOO
00000
0000
00-

00000 00000 00000 00000 O W 0 0 00000 moo

00000 00000 00300 00030 00000 90000 0-0
00000 OOOOO 00000 00000 00000 00000 0000
00000 00000 00000 00000 00000 00000 0 0 O N
9
C
??YE 33%?
COO000 00000 00000
????. ".9'.9?3 33??? ?'.?3? 9'7??
00000 00000 00000 00011)

UDWU ownee OIQON -none ~ 0 0 0 0CMI,U 0000

W O V 9 0 9 9-90) O l O O Q U C O C N Z Q O O O UWI11)OUl Q-90
u
..................................
r%VOIW UOQOD Nn11)U14 W O - N U O w - OO-NO -DO
L mm-011) me-n.n ~ ~ 1 m 1 o0t ~ o~ o o o oooooo nn~o
09J03 U->V.V -00- O.7Ua- O W a O OaOO- UtOO
4 C
a 00000 O O W O 00000 0 0 0 0 0 00000 00000 0000

OONO
0000

....
m-0
9010
0000
0000

N- 4-0
11)9U t U 0000

.....
0011)CC 0000
cnon* UUOO
OOOOO
O W 0 0
?Y.9
0000
ONbsO 00111)
00-C 4rlll11)

.....
DW--
0-,
roo00
00000
....
CC11)N
2!2zg
0000
Z
*
0
4 0 -4
OI+.IvoI? OC-
O O M I OS9.2-
44, I ? C O )
QOOOO QOO00 001)Oo
-d
*S-
02s-
-99-
*O.rD
O S I O O -DOD
ul*OFIN O O I L n I I - I(YOIII)
0-
x ...................a
n
C
Z
W
a
0
3
3
c
W

xx-
21-2
W

ww-=*
mats+-
0
C
Szaa IS
0333 00
z x z d s A Y ~ S D ax
S 3 a a I Z d3I15 3 3 - S
N'PI~ mnaa-
-@322 * - 1 S 1
Sooaa om--
a

-=co
cainxx arccx -c*a
t-m
*I 0
003 %
X

13.uId3
~ N Y ' O
e
-SS-
a - =
rc-
~a a ~ ~ ~
a m d a .&41'&T.
-2-
-a-
=ws&
w-gc~
cch
5
a- Ja
-
J-
'&r..C
$

ZS
m 9 r o o 1-0-
**Coo no-
r-noo - = w o o 0
aPB-00 no-
no000
c-
-9000
wmn
M-C-0
000ac
001m

.....
ocom
OOONO
oowo
IIIII
 q OIDW
-nsm
n n n o e n r r r a r -QYI
8-9 NIIO~ s909
gcroo
nrnro n m o o o CIO~IYN
oe-0- n
- etnal-0
lor-
L ulm- o n m u r u o n e -un 4-9 uaerotu ON- oOr09
,
n

5
omo-
000,
-8000
0-0
O m 0 4 OOCOO
,000 ,NO0 o v;;c;,'f d;&G
==Y!!!!!!!!xE?~
&AS ;,'s;
g 42-5
00000 00000
t I
:?
' 2:4
II
439::
OOOOO
II
33z:
00000
I
,Z
0

/
n
0
C
C

3
*
2
x
n
I
nrr-
000.-

8 ss494 S?.SS
n 00000
N I 1
o
-NN
-
n n s v ' n -09--
-e
0000-
~s r o c aaoor
9.-D-
-I0
o 3 w o ooocuo 00000 00000
~
r N N o r moo90
1
O N 0 4

sssss sssss
00000 00000
I 1 Ill
00000
1
~

o
5
0
nrrmn
cV-)r,
4
c-.
I
-
-OD-
-,3r

p
11
oeono
*n.nrw
%?3S
%SSZ
-.*.*.
r rrod e w e
'7 ' 7

a
1
b
3 P
0 .c
4
f
s3
.
f
(I)
g
N
n
-
Z
rl
3
(1
rl
II)
m
3 I W r
Ulw
Z
0
>
x
f w
z L%tS 5.
a a
" O I -
023 3:=
C.2
9

k:&4gp2
3 -203
>-,t
-2-
s 4 4 332%
a-Ze
H-,z
-ax0
2
:
s
+
03=HA
=A= C*C.
& E 4 3 gg&
 CC01oQ
.....
*CON
OIOOO
QINOI
O*N N
4 4 4

CC99N 00- 0-IQC -4

ul9O-v9 C L P I O I e*-.nU) -9-9
taana ccton o-n-I coao.r
0 4 0 0 1 00010- C-NdU NO-
....................
00nn~ nnnor-
010101Ca 01001-
00000 00000
-.nn.n PCOCQ
OOCCO O N 4 0 1
00000 00000

N ~ N O I UINOO nano- 01-00

ry9.n- 9-409 0-0 OOONI
7-23? 7?77. %-.? 7-.%
tmr-w -om mne-a wonu*n
8.5. REFERENCES
1. Wagner and Evans, 1933. Symmetrical Components. McGraw-Hill, New York.
2. A. Brameller, 1972. 'Three-phase short-circuit calculation using digital computer.'
Internal Report, University of Manchester Institute of Science and Technology.
3. A. Brameller, 1972. 'User manual for unbalanced fault analysis.' Internal Report,
University of Manchester Institute of Science and Technology.
4. J. Preece, 1975. 'Symmetrical components fault studies.' MSc. Dissertation, University
of Manchester Institute of Science and Technology.
5. C. B. Lake, 1978. 'A computer program for a.c. fault studies in a.c.-d.c. systems.' M.E.
thesis, University of Canterbury, New Zealand.
6. B. Stott and E. Hobson, 1971. 'Solution of large power-system networks by ordered
elimination: a comparison of ordering schemes', Proc. IEE, 118 (I), 125-134.
7. K. Zollenkopf, 1970. 'Bifactorization-basic computational algorithm and pro-
gramming techniques.' Conference on Large Sets of Sparse Linear Equations, Oxford,
75-96.
8. P. M. Anderson, 1973. Analysis of Faulted Power Systems. Iowa State University Press.
 Power System Stability-
Basic Model
9.1. INTRODUCTION
The stability of a power system following some predetermined operating
condition is a dynamic problem and requires more elaborate plant component
models than the ones discussed in previous chapters. It is normally assumed that
prior to the dynamic analysis, the system is operating in the steady state and that
a load-flow solution is available.
Two types of stability studies are normally carried out. The subsequent
recovery from a sudden large disturbance is referred to as 'Transient Stability7
and the solution is obtained in the time domain. The period under investigation
can vary from a fraction of a second, when first swing stability is being
determined, to over ten seconds when multiple swing stability must be examined.
The term 'Dynamic Stability7is used to describe the long-time response of a
system to small disturbances or badly set automatic controls. The problem can be
solved either in the time domain or in the frequency domain. In this book,
dynamic stability is treated as an extension of transient stability and is thus solved
in the time domain. Such extension normally requires modification of some plant
component models and often the introduction of new models, but because of the
smaller perturbations and longer study duration the small time constant effects
can be ignored.
Consideration is given in this chapter to the dynamic modelling of a power
system containing synchronous machines and basic loads. More advanced
synchronous machine models as well as other power-system components, such as
induction motors and a.c.-d.c. converters, are considered in Chapter 10.

The form of the equations

To a greater or lesser extent, all system variables require time to respond to any
change in operating conditions and a large set of differential equations can be
written to determine this response. This is impractical, however, and many
assumptions must be made to simplify the system model. The assumptions made
depend on the problem being investigated and no clear definitive model exists.
 A major problem with a time domain solution is the 'stflness' of the system
(Appendix I). That is, the time constants associated with the system variables vary
enormously. When only synchronous machines are being considered, rotor swing
stability is the principal concern. The main time constants associated with the
rotor are of the order of 1 to 10 seconds. The form of the solution is dominated by
time constants of this order and smaller or greater time constants have less
significance.
The whole of the a.c. transmission network responds rapidly to configurational
changes as well as loading changes. The time constants associated with the
network variables are extremely small and can be considered to be zero without
significant loss of accuracy. Similarly the synchronous machine stator time
constants may be taken as zero. The relevant differential equations for these
rapidly changing variables are transformed into algebraic equations.
When the time constant is large or the disturbance is such that the variable will
not change greatly, the time constant may be regarded as infinite, that is the
variable becomes a constant. Excitation voltage or mechanical power to the
synchronous machine may often be treated as constant in short duration studies
without appreciable loss of accuracy. Depending on how the computer program
is written variables which become constant may be treated by either:
(1) retaining the differential equation but assigning a very large value to the
relevant time constant;
(2) removing the differential equation.
For flexibility both methods are usually incorporated in a program.
A system which after these initial simplifications contains a differential
variables, contained in the vector Y,, and P algebraic variables, contained in the
vector X,, may be described by the matrix equations:

d
where p denotes the differential operator -.
dt

Frames of reference
The choice of axes, or frame of reference, in which the system equations are
formulated is of great importance, as it influences the analysis.
For synchronous machines, the most appropriate frame of reference is one
which is attached to the rotor, i.e. it rotates at the same speed as the rotor. The
main advantage of this choice is that the coefficients of the equations developed
for the synchronous machine are not time-dependent. The major axis of this
frame of reference is taken as the rotor pole or 'direct axis'. The second axis lies
90 degree (electrical) from each pole and is referred to as the 'quadrature axis'.
In the dynamic state, each synchronous machine is rotating independently
from each other and transforming between synchronous machine frames through
the network is difficult. This is overcome by choosing an independent frame of
reference for the network and transforming between this frame and the
synchronous machine frames at the machine terminals. The most obvious
choice for the network is a frame of reference which rotates at synchronous speed.
The two axes are obtained from the initial steady-state load-flow slack busbar.
Although the network frame is rotating synchronously, this does not stop each
nodal voltage or branch current from having an independent frequency during
the dynamic analysis.

9.2. SYNCHRONOUS MACHINES-BASIC MODELS

Mechanical equations
The mechanical equations of a synchronous machine are very well estab-
li~hed(').(~)
and need be only briefly outlined. Three basic assumptions are made
in deriving the equations:
Machine rotor speed does not vary greatly from synchronous speed
(1.0 P.u.).
Machine rotational power losses due to windage and fric'tion are ignored.
Mechanical shaft power is smooth, that is the shaft power is constant
except for the results of speed governor action.
Assumption 1 allows per unit power to be equated with per unit torque. From
Assumption 2, the accelerating power of the machine (Pa) is the difference
between the shaft power (Pm) as supplied by the prime mover or absorbed by the
load and the electrical power (Pe). The acceleration (a) is thus:

where Mg is the angular momentum.

The acceleration is independent of any constant speed frame of reference and it
is convenient to choose a synchronously rotating frame to define the rotor angle
(6). Thus:
d26 - (Pm - Pe)
--
dt2 Mg
The angular momentum may be further defined by the inertia constant Hg
(measured in MWSIMVA) which is relatively constant regardless of the size of
the machine, i.e.:

where fo is the system base frequency.

Eddy currents induced in the rotor iron or in the damping windings produce
torques which oppose the motion of the rotor relative to the synchronous speed.
 209
A decelerati ower ca ntroduced into the m~echanicalequations to
account for thisdamping, giving

The damping coefficient (Da), measured in watts/rad/sec, has been largely

superseded by a synchronous machine model which includes the subtransient
effect of the damper windings in the electrical equations, but it is still used in some
programs.
Two single-order ordinary differential equations may now be written to
describe the mechanical motion of the synchronous machine, i.e.
1
po = -(Pm - Pe - Da(o - 2nfo))
Mg

Electrical equations
The derivation of equations to account for flux changes in a synchronous
machine has been given by C ~ n c o r d i a 'and
~ ) Kimba~-k.(~)
A brief outline only will
be given in this section, so that various electrical quantities may be defined and
phasor diagrams'constructed. The approximations made in the derivation are as
follows:
( 1 ) The rotor speed is always sufficiently near 1.0 p.u. that it may be considered
a constant.
(2) All inductances defined in this section are independent of current. The
effects due to saturation of iron are considered in Chapter 10.
(3) Machine winding inductances can be represented as constants plus
sinusoidal harmonics of rotor angle.
(4) Distributed windings may be represented as concentrated windings.
(5) The machine may be represented by a voltage behind an impedance.
(6) There are no hysteresis losses in the iron, and eddy currents are only
accounted for by equivalent windings on the rotor.
(7) Leakage reactance only exists in the stator.
Using these assumptions, classical theory permits the construction of a model for
the synchronous machine in the steady-state, transient and subtransient states.
The per unit system adopted is normalized to eliminate factors of &a, n
and turns ratio, although the term 'proportional' should be used instead of 'equal'
when comparing quantities. Note that one p.u. field voltage produces 1.0 p.u. field
current and 1.0 p.u. open-circuit terminal voltage at rated speed.
STEADY STATE EQUATIONS
Figure 9.1 shows the flux and voltage phasor diagram for a cylindrical rotor
synchronous machine in which all saturation effects are ignored. The flux Ff
 Direct axis

Quadrature
axls

Fig. 9.1. Phasor diagram of a cylindrical rotor synchronous machine in the

steady state

is proportional to the field current If and the applied field voltage and it acts in the
direct axis of the machine. The stator open-circuit terminal voltage Ei is
proportional to Ff but lies on the quadrature axis. The voltage Ei is also
proportional to the applied field voltage and may be referred to as Ef.
When the synchronous machine is loaded, a flux F proportional to and in
phase with the stator current I is produced which, when added vectorially to the
field flux Ff, gives an effective flux Fe. The effective internal stator voltage El is
due to Fe and lags it by 90 degrees. The terminal voltage V is found from this
voltage El by considering the volt drops due to the leakage reactance XI and
armature resistance Ra. By similar triangles, the difference between Ef and El is in
phase with the I X l volt drop and is proportional to I . Therefore the voltage
difference may be treated as a volt drop across an armature reactance X a . The
sum of X a and XI is termed the synchronous reactance.
For the salient pole synchronous machine the phasor diagram is more
complex. Because the rotor is symmetrical about both the d and q axes it is
convenient to resolve many phasor quantities into components in these axes.
The stator current may be treated in this manner. Although F , will be
proportional to I, and F, will be proportional to I,, because the iron paths in the
two axes are different, the total armature reaction flux F will not be proportional
to I nor necessarily be in phase with it. Retaining our earlier normalizing
assumptions, it may be assumed that the proportionality between I , and F , is
unity but the proportionality between I, and F, is less than unity and is a function
of the saliency.
In Fig. 9.2 the phasor diagram of the salient pole synchronous machine is
shown. Note that the d and q axes armature reactances have been developed as in
the cylindrical rotor case. From these, direct and quadrature synchronous
Direct axis

\A ,,
Quadrature
f axis

Fig. 9.2. Phasor diagram of a salient pole synchronous machine in the steady state

reactances (X, and X,) can be established, i.e.

where Vd and V, are the axial components of the terminal voltage V.

In steady-state conditions it is quite acceptable to use as the machine model,
the field voltage Ef or the voltage equivalent to field current Ei behind the
synchronous reactances. In these circumstances the rotor position (quadrature
axis) with respect to the synchronously rotating frame of reference is given by the
angular position of Ef .
Only the salient pole machine will now be considered, as the cylindrical rotor
model may be regarded as a special case of a salient machine (X,= X,).
TRANSIENT EQUATIONS
For faster changes in the conditions external to the synchronous machine, the
above model is no longer suitable. Due to the 'inertia' of the flux linkages these
changes can not be reflected throughout the whole of the model immediately. It is
therefore necessary to create new fictitious voltages Ei and Ei which represent the
flux linkages of the rotor windings. These transient voltages can be shown to exist
behind the transient reactances Xi and Xi.
 The voltage Ei should now be considered as the sum of two voltages, Ed and E,,
and is the voltage behind synchronous reactance. In the previous section, where
steady state was considered, current flowed only in the field winding and, hence,
in that case, Ed = 0 and E, = E,.
Where it is necessary to allow the rotor flux linkages to change with time, the
following ordinary differential equations are used:

The phasor diagram of the machine operating in the transient state is shown in
Fig. 9.3.

\
Fig. 9.3. Phasor diagram of a synchronous machine in the transient state

SUBTRANSIENT EQUATIONS
Either deliberately, as in the case of damper windings, or unavoidably, other
circuits exist in the rotor. These circuits are taken into account if a more exact
model is required. The reactances and time constants involved are small and can
often be justifiably ignored. When required, the development of these equations is
identical to that for transients and yields:
E: - V, = RaI, - X,"Id (9.2.15)
E,"- Vd = RaId + X i I , (9.2.16)
pEi = (Ei + (XI;- X i ) I d - Ei)/T;, (9.2.17)
pE," = (El;- ( X i - X,")I, - E:)/Tlj, (9.2.18)
The equations are developed assuming that the transient time constants are
large compared with the subtransient time constants. A phasor diagram of the
synchronous machine operating in the subtransient state is shown in Fig. 9.4. It
 Fig. 9.4. Phasor diagram of a synchronous machine in the subtransient state

should be noted that equations (9.2.11) and (9.2.12) are now true only in the
steady-state mode of operation, although once subtransient effects have decayed,
the error will be small.
MACHINE MODELS

It is possible to extend the model beyond subtransient level but this is seldom
done in multi-machine programs. investigation^(^) using a generator model with
up to seven rotor windings, have shown that using the standard machine data the
more complex models do not necessarily give more accurate results. However,
improved results can be obtained if the data, especially the time constants are
suitably modified.
The most convenient method of treating synchronous machines of differing
complexity is to allow each machine the maximum possible number of equations
and then let the actual model used be determined automatically according to the
data presented.
Five models are thus possible for a four-winding rotor.
Model 1-constant voltage magnitude behind d-axis transient reactance (Xi)
requiring no differential equations. Only the algebraic equations (9.2.11) and
(9.2.12) are used.
Model 2-D-axis transient effects requiring one differential equation (pEh).
Equations (9.2.1I), (9.2.12) and (9.2.13) are used.
Model 3-0- and q-axis transient effects requiring two differential equations
(pEb and pE;). Equations (9.2.1I), (9.2.12), (9.2.13) and (9.2.14) are used.
Model 4--0- and q-axis subtransient effects requiring three differential
equations (pEi, pE," and pEi). Equations (9.2.13), (9.2.15), (9.2.16), (9.2.17) and:
 are used. This last equation is merely equation (9.2.14) with modified primes.
Whether it is a subtransient or transient equation is open to argument.
Model 5-D- and q-axis subtransient effects requiring four differential
equations (pE;, pE;, pEi and pE;). Equations (9.2.13),(9.2.14), (9.2.15), (9.2.16),
(9.2.17) and (9.2.18) are used.
The mechanical equations (9.2.5) and (9.2.6) are also required to be solved for
all these models.
Groups of synchronous'machinesor parts of the system may be represented by
a single synchronous machine model. An infinite busbar, representing a large stiff
system, may be similarly modelled as a single machine represented by model 1,
with the simplification that the mechanical equations (9.2.5) and (9.2.6) are not
required. This sixth model is thus defined as:
Model 0-Infinite machine-constant voltage (phase and magnitude) behind d-
axis transient reactance (Xi). Only equations (9.2.1 l ) and (9.2.12) are used.

9.3. SYNCHRONOUS MACHINE AUTOMATIC CONTROLLERS

For dynamic power system simulations of 1 second or longer duration, it is
necessary to include the effects of the machine controllers, at least for the machine
most affected by the disturbance. Moreover, controller representation is
becoming necessary, even for first swing stability, with systems being operated at
their limits with near critical fault clearing times.
The two principal controllers of a turbine-generator set are the automatic-
voltage regulator (AVR) and the speed governor. The AVR model consists of
voltage sensing equipment, comparators and amplifiers controlling a synch-
ronous machine which can be generating or motoring. The speed governor may
be considered to have similar equipment but in addition it is necessary to take the
turbine into account.

Automatic voltage regulators

Many different AVR models have been developed to represent the various types
used in a power system. The application of such models is dficult and a better
approach is to develop a single general purpose AVR model, on a similar basis to
the synchronous machine model. The model can then revert to any desired type
by using the correct data. The IEEE defined several AVR types,'6) the main two of
which (Type 1 and Type 2) are shown in Fig. 9.5.
A composite model of these two AVR types can be constructed. This model
may also include a secondary signal which can be taken from any source, but
usually either machine rotor speed deviation from synchronous speed or rate of
change of machine output power. This model is shown in Fig. 9.6 and has been
found to be satisfactory for all the systems studied so far. It is acknowledged that
other AVR models may be necessary for specific studies.
In many systems studied, the amount of data available for an AVR model is
 S e = f ( E f ) +

Saturating
Regulator amplifter exciter
1 Ef

busbar voltage voltage

I
Other
sig&ls
- Kfp
(1+T/.p)
Feedback loop

voltage

Controlled
voltage

Other
signals Feedback loop

Fig. 9.5. Block diagrams for two commonly used AVR models. (6) (a) IEEE Type 1
AVR model; (b) IEEE Type 2 AVR model. (01982 IEEE)

Vt
Co$olled =
1 a,,' + -
Z -- 1
( U ~ T W') T e l d
N

busbar voltage
voltage -

signal

Fig. 9.6. Block diagram of a composite automatic voltage regulator model

quite small. The composite model can degenerate into a very simple model easily
by defaulting time constants to zero and gains to either zero, unity or an
extremely large value depending on their position.
The equations for the AVR model shown in Fig. 9.6 are as follows:
216
subject to

and

pVf b = (Vd - V f b ) / T , (9.3.5)

where Se =f ( E f )
V g = E f [unless IEEE Type 2 when V g = Val (9.3.10)
Vaux= A predefined signal
The IEEE(6)recommends that Se be specified at maximum field voltage (Se,,)
and at 0.75 of maximum field voltage (Seo.7,,ax).From this Se may be determined
for any value of field voltage by either linear interpolation or by fitting a
quadratic. Where linear interpolation is used, equation (9.3.9)may by transfor-
med to:
V e = ( k , .Ef - k2)Ef (9.3.11 )
where

A means of modelling lead-lag circuits such as those in the regulator amplifier, the
stabilizing loop and the auxiliary signal circuits is given at the end of this section.
Despite the advantages of one composite AVR model, if there are a great many
AVRs to be modelled most of which have simple characteristics then it is better to
make two models. One model, which contains only the commonly used parts of
the composite model can then be dimensioned for all AVRs. The other model,
which contains only the less commonly used parts of the composite model can be
quite small dimensionally. A connection vector is all that is necessary to
interconnect the two models whenever necessary.
Speed governors
For speed governors, as with AVRs, a composite model which can be reduced to
any desired level is the most satisfactory. The speed governor models recom-
mended by the IEEE(7)are shown in Fig. 9.7. It will be noticed that if limits are
not exceeded, the two models are identical. The difference is due to the

Specifled Ps
power setting 7
4 v

speed power setting

277 fo Ref. speed

(a)

Specified
power sett~ng 7"
M,~,,,,
speed power setting
277 f o b Ref. Speed
(b)
Fig. 9.7. Typical models of speed governors and valves. (7) (a) Thermal governor
and valve; (b) Hydro governor and valve. (01982 IEEE)

( 1 t T1.p) ( 1 t T3.p)
Power to
I turbine

Fig. 9.8. Generalized model of a speed governor and valve

assumption that, in a hydro governor, gate servo and gate positions are the same.
One model can be used for the governors of both turbines provided that the limits
are either internal or external to the second transfer function block of Fig. 9.8.
Also, very little extra effort is required to divorce the governor from the actual
turbine power and keep it instead as a function of valve position.
The equations of the speed governor shown in Fig. 9.8 are:

The valvelgate position setting (Gv) is subject to opening and closing rate limits
(omaxand cmX respectively) and to physical travel limits so that:

The valve equation is:

Pgv = G v .Pb (9.3.16)
For thermal turbines, where a boiler is modelled, in the steady state, Pb will be
the actual power delivered and Gs will be unity, i.e. the valve will be fully open. If a
boiler is not modelled or a hydro turbine is being controlled then, in the steady
state, Pb will be the maximum output from the boiler or water system (i.e.
maximum turbine mechanical power output) and Gv, and hence Gs, will be such
that Pgv is the actual mechanical power output of the turbine.
This method of modelling a valve has the advantage that nonlinearities
between valve position and power can be easily included and also the operation
of the governor and valve can be readily interpreted.
For a hydro governor where the limits are external, the model is as given in
equations (9.3.12)to (9.3.16) but for a thermal governor, G2 is reset after the valve
limits are applied to be:
G2(,,,, = Gv - Gs (thermal governor only) (9.3.17)

Hydro and thermal turbines

This section is restricted to the modelling of simple turbines only. Compound
thermal turbines may require a detailed model, as given in Chapter 10, but, for
stability studies of only 1 or 2 seconds duration, the effect of all but the high
pressure (HP) turbine can usually be ignored. The time constant associated with
the steam entrained between the H P turbine outlet and the I P or LP turbine inlet
is usually very large (greater than 5 sec.) and the output from all turbines other
than the H P turbine may be treated as constant.
Simple linear models of hydro and thermal turbines are shown in Fig. 9.9. The
hydro-turbine model includes the penstock which gives the characteristic lead-

Power delivered 1 '''"'PII Mechan~calpower

through gate to generator

pw
Power dehvered (1tT4p)
through valve
PI
I
Mechanical power
to generator
/P and L P
(b) turbine power

Fig. 9.9. Simple linear models of turbines.

(a) Hydro turbine; (b) Thermal turbine
lag response of this type of turbine. The model is generally sufficient for all hydro
turbines and, from Fig. 9.9, the differential equation for the mechanical power
output (Pm)of the turbine is:

with T4 = 0.5Tw as a further close approximation.

For the thermal turbine using Fig. 9.9(b)this equation is:

with K1 representing the fraction of power delivered by the H P turbine. For

simple turbines K1 is thus unity. For compound turbines, the power (PI)from the
I P and LP turbines is obtained from:

here Pm, is the initial steady-state mechanical power. Note that for this simple
model, the initial value of Pgv is PmIK 1, even though all the steam passes through
the valve.
Provided that the H P valve does not close fully, then, rather than inject the
power from the I P and L P turbines as shown, it is easier to treat it as a simple
turbine (PI = 0 ) but with the speed regulation modified by:

Modelling lead-lag circuits

Lead-lag circuits may present a problem depending on the integration scheme
adopted. Where the differential equations are not used directly and the
derivatives are not explicitly calculated, the following can be used to convert the
model into a more acceptable form.

Fig. 9.10. Typical lead-lag circuit block

diagram

For the circuit shown in block diagram form in Fig. 9.10, the equation is:

K(1+ T2p)
vout =
+
( 1 T l p ) Vin
This can be transformed to:
+
vout =-(
K . T 2 (TlIT2) T I - p
T1 1 T1.p + ) 'in
and then to:

which can be represented by the block diagram in Fig. 9.1 1, and is a lag circuit in
parallel with a gain.
It is important to remember that the time constant TI, must be nonzero even if
the integration method can accommodate zero-time constants.

Fig. 9.11. Modified block diagram of a lead-lag

circuit

9.4. LOADS
Early transient-stability studies were concerned primarily with generator
stability, and little importance was attached to loads. In the two-machine
problem for example, the remainder of the system, generators and loads, were
represented by an infinite busbar. A great deal of attention has been given to load
modelling since then.
Much of the domestic load and some industrial load consist of heating and
lighting, especially in the winter, and in early load models these were considered
as constant impedances. Rotating equipment was often modelled as a simple
form of synchronous machine and composite loads were simulated by a mixture
of these two types of load.
A lot of work has gone into the development of more accurate load models.
These include some complex models of specific large loads which are considered
in the next chapter. Most loads, however, consist of a large quantity of diverse
equipment of varying levels and composition and some equivalent model is
necessary.
A general load character is ti^(^) may be adopted such that the MVA loading at
a particular busbar is a function of voltage (V) and frequency (f):

where K p and Kq are constants which depend upon the nominal value of the
variables P and Q.
 Fig. 9.12. Characteristic of different load mo-
dels. (a) Active and reactive power against
voltage; (b) Current against voltage

Static loads are relatively unaffected by frequency changes, i.e. p f = q f = 0, and

with constant impedance loads pv = qv = 2.
The importance of accurate load models has been demonstrated by Dandeno
and Kundur") when considering voltage sensitive loads. Figure 9.12 de-
monstrates the power and current characteristics of constant power, constant
current and constant impedance loads. Berg(g)has identified the characteristic
load parameters for various homogeneous loads and these are given in Table 9.1.
These characteristics may be combined to give the overall load characteristic at a
busbar. For example, a group of n homogeneous loads, each with a characteristic
of pvj and a nominal power of Pj may be combined to give an overall
characteristic of:
n / n

The other three overall characteristics may be similarly determined.

 Table 9.1. Typical values of characteristic load parameted9'

Load
Filament lamp
Fluorescent lamp
Heater
Induction motor half load
Induction motor full load
Reduction furnace
Aluminium plant

Low-voltage problems
When the load parameters pv and qv are less than or equal to unity, a problem can
occur when the voltage drops to a low value. As the voltage magnitude decreases,
the current magnitude does not decrease. In the limiting case with zero-voltage
magnitude, a load current flows which is clearly irrational, given the nondynamic
nature of the load model. From a purely practical point of view, then the load
characteristics are only valid for a small voltage deviation from nominal. Further,
if the voltage is small, small errors in magnitude and phase produce large errors in
current magnitude and phase. This results in loss of accuracy and with iterative
solution methods poor convergence or divergence.
These effects can be overcome by using a constant impedance characteristic to
represent loads where the voltage is below some predefined value, for example 0.8
p.u.

9.5. THE TRANSMISSION NETWORK

It is usual to represent the static equipment which constitute the transmission
system by lumped 'equivalent pi' parameters independent of the changes occurring
in the generating and load equipment. This representation is used for multi-
machine stability programs because the inclusion of time varying parameters
would cause enormous computational problems. Moreover, frequency, which is
the most obvious variable in the network, usually varies by only a small amount
and thus, the errors involved are small. Also, the rates of change of network
variables are assumed to be infinite which avoids the introduction of differential
equations into the network solution.
The transmission network can thus be represented in the same manner as in the
load-flow or short-circuit programs, that is by a square complex admittance
matrix.
The behaviour of the network is described by the matrix equation:
['inj] = CYICJ'I (9.5.1)
where [Iinj]is the vector of injected currents into the network due to generators
and loads and [qis the vector of nodal voltages.
Any loads represented by constant impedances may be directly included in the
network admittance matrix with the injected currents due to these loads set to
zero. Their effect is thus accounted for directly by the network solution.

9.6. OVERALL SYSTEM REPRESENTATION

Two alternative solution methods are possible. The preferred method uses the
nodal matrix approach, while the alternative is the mesh matrix method.
Matrix reduction techniques can be used with both methods if specific network
information is not required, but this gives little advantage as the sparsity of the
reduced matrix is usually very much less.

Mesh matrix method

In this method, the system-loading components are treated as Thevenin
equivalents of voltages behind impedances. The network is increased in size to
include these impedances and the mesh impedance matrix of the increased
network is created. This is then inverted or the factorized form of the inverse
determined.
The solution process is as follows:
(1) Calculate the Thevenin voltages of the system loading components by
solving the relevant differential and algebraic equations.
(2) Determine the network currents using the Y matrix or factors. As the
network current around a mesh containing the Thevenin voltage is the
loading current this may affect the Thevenin voltage in which case an
iterative process will be required.

Nodal matrix method

In this method, all network loading components are converted into Norton
equivalents of injected currents in parallel with admittance. The admittances can
be included in the network admittance matrix to form a modified admittance
matrix which is then inverted, or preferably factorized by some technique so that
solution at each stage is straightforward.
The following solution process applies:
(1) For each network loading component, determine the injected currents into
the modified admittance matrix by solving the relevant differential and
algebraic equations.
(2) Determine network voltages from the injected currents using the 2-matrix
or factors.
As the network voltages affect the loading components, an iterative process is
often required, although good approximation^'^ can be used to avoid this.
With the Nodal Matrix Method, busbar voltages are available directly and
branch currents can be calculated if necessary while with the mesh matrix
method, mesh currents are available directly and busbar voltages and branch
currents must be calculated if necessary.
Although much work has been spent on the systematic construction of the
mesh impedance matrix, the nodal admittance matrix is easier to construct and
has gained wide acceptance in load-flow and fault analysis. For this reason, the
remainder of this section will consider the nodal matrix method.

Synchronous machine representation in the network

The equations representing a synchronous machine, as defined in Section 9.2, are
given in the form of Thevenin voltages behind impedances. This must be modified
to a current in parallel with an admittance by use of Norton's theorem. The
admittance of the machine thus formed may be added to the shunt admittance of
the machine busbar and treated as a network parameter. The vector [Iinj]in
equation (9.5.1) thus contains the Norton equivalent currents of the synchronous
machines.
The synchronous machine equations are written in a frame of reference
rotating with its own rotor. The real and imaginary components of the network
equations, as given in Fig. 9.13, are obtained from the following transformation:
Netwprk

Machine
A lmaglnary axis
direct axis
---- ----- v Machine
quadrature axis

= Network real axis

Fig. 9.1 3. Synchronous machine and network frames of

reference

VLW.-qI
This transformation is equally valid for currents as is the reverse transformation:

-sin 6 cos 6
 When saliency exists, the values of X i and Xi used in equations (9.2.15) and
(9.2.16) and/or X i and X i used in equations (9.2.11) and (9.2.12) are different.
Therefore, the Norton shunt admittance will have a different value in each axis
and when transformed into the network frame of reference, will have time varying
components. However, a constant admittance can be used, provided that the
injected current is suitably modified to retain the accuracy of the Norton
equivalent.(lO)This approach can be justified by comparing the two circuits of
Fig. 9.14 in which Yt is a time varying admittance, whereas yo is fixed.

(bJ
Fig. 9.14. Method of representing synchronous
machines in the network. (a) Norton equivalent
circuit; (b) Modified equivalent circuit

At any time t, the Norton equivalent of the machine is illustrated in Fig. 9.14(a),
but the use of a fixed admittance results in the modified circuit of Fig. 9.1qb).
The machine current is:

and hence

where iadj accounts for the fact that the apparent current source is not accurate in
this case.
The injected current into the network which includes Yo is given by:

where
- -
Zunadj = YOEn
 A suitable value for yo is found by using the mean of direct and quadrature
admittances, i.e.

where
Xdq= +(Xi + Xi)
The unadjusted value of current injected into the busbar is:

(9.6.6)
The adjusting current is not affected by rotor position in the machine frame of
reference but it is when considered in the network frame. From equation (9.6.3)
and also equations (9.2.15) and (9.2.16):

and transforming

The total nodal injected current is therefore

Load representation in the network

To be suitable for representation in the overall solution method, loads must be
transformed into injected currents into the transmission network from which the
terminal voltages can be calculated. A Norton equivalent model of each load
must therefore be created. In a similar way to that adopted for synchronous
machines, the Norton admittance may be included directly in the network
admittance matrix.
 A constant impedance load is therefore totally included in the network
admittance matrix and its injected current is zero. This representation is
extremely simple to implement, causes no computational problems and improves
the accuracy of the network solution by strengthening the diagonal elements in
the admittance matrix.
Nonimpedance loads may be treated similarly. In this case, the steady-state
values of voltage and complex power obtained from the load flow are used to
obtain a steady-state equivalent admittance (Yo)which is included in the network
admittance matrix [a. During the stability run, each load is solved sequentially
along with the generators, etc. to obtain a new admittance (P), i.e.:

The current injected into the network thus represents the deviation of the load
characteristic from an impedance characteristic.

Fig. 9.15. Load and injected currents for a

constant power type load with low-voltage
adjustment. (a) Load current; (b) Injected
Current
 By converting the load characteristic to that of a constant impedance, when the
voltage drops below some predetermined value (V,,,), as described in Section
9.4, the injected current is kept relatively small. An example of a load
characteristic and its corresponding injected current is shown in Fig. 9.15.
In an alternative model the low-voltage impedance is added to the network and
the injected current compensates for the deviation from the actual characteristic.
In this case, there is a nonzero injected current in the initial steady-state operating
condition.

System faults and switching

In general most power-system disturbances to be studied will be caused by
changes in the network. These changes will normally be caused by faults and
subsequent switching action but occasionally the effect of branch or machine
switching will be considered.

FAULTS
Although faults can occur anywhere in the system, it is much easier com-
putationally to apply a fault to a busbar. In this case, only the shunt admittance at
the busbar need be changed, that is, a modification to the relevant self-admittance
of the Y matrix. Faults on branches require the construction of a dummy busbar
at the fault location and suitable modification of the branch data unless the
distance between the fault position and the nearest busbar is small enough to be
ignored.
The worst case is a three-phase zero-impedance fault and this involves placing
in infinite admittance in parallel with the existing shunt admittance. In practice, a
nonzero but sufficiently low-fault impedance is used so that the busbar voltage is
effectively brought to zero. This is necessary to meet the requirements of the
numerical solution method.
The application or removal of a fault at an existing busbar does not affect the
topology of the network and where the solution method is based on sparsity
exploiting ordered elimination, the ordering remains unchanged and only the
factors required for the forward and backward substitution need be modified.
Alternatively the factors can remain constant and diakoptical techniques(") can
be used to account for the network change.

BRANCH SWITCHING
Branch switching can easily be carried out by either modifying the relevant
mutual- and self-admittances of the Y matrix or using diakoptical techniques. In
either case, the topology of the network can remain unchanged as an open branch
is merely one with zero admittance. While this does not fully exploit sparsity, the
gain in computation time by not reordering exceeds the loss by retaining zero
elements, in almost all cases.
The only exception is the case of a branch switched into a network where no
interconnection existed prior to that event. In this case, either diakoptical or
reordering techniques become necessary. To avoid this problem, a dummy*
branch may be included with the steady-state data of sufficiently high impedance
that the power flow is negligible under all conditions, or alternatively, the branch
resistance may be set negative to represent an initial open circuit. A negative
branch reactance should not be used as this is a valid parameter where a branch
contains series capacitors.
Where a fault occurs on a branch but very close to a busbar, nonunit protection
at the near busbar will normally operate before that at the remote end. Therefore,
there will be a period when the fault is still being supplied from the remote end.
There are two methods of accounting for this type of fault.
The simplest method only requires data manipulation. The fault is initially
assumed to exist at the local busbar rather than on the branch. When the specified
time for the protection and local circuit bareaker to operate has elapsed, the
fault is removed and the branch on which the fault is assumed to exist is opened.
Simultaneously, the fault is applied at the remote busbar, but in this case, with the
fault impedance increased by the faulted branch impedance, similarly the fault is
maintained until the time specified for the protection and remote circuit breaker
to operate has elapsed.
The second method is generally more involved but it is better when protection
schemes are modelled. In this case, a dummy busbar is located at the fault
position, even though it is close to the local busbar and a branch with a very small
impedance is inserted between the dummy busbar and the local busbar. The
faulted branch then connects the dummy busbar to the remote busbar and the
branch shunt susceptance originally associated with the local busbar is
transferred to the dummy busbar. This may all be done computationally at the
time when the fault is being specified. The two branches can now be controlled
independently by suitable protection systems. An advantage of this scheme is that
the fault duration need not be specified as part of the input data. Opening both
branches effectively isolates the fault, which can remain permanently attached to
the dummy busbar, or if auto-reclosing is required, it can be removed
automatically after a suitable deionization period.
The second method will give problems if the network is not being solved by a
direct method. During the iterative solution of the network, slight voltage errors
will cause large currents to flow through a branch with a very small impedance.
This will slow convergence and in extreme cases will cause divergence. With a
direct method, based on ordered elimination, an exact solution of the busbar
voltages is obtained for the injected currents specified at that particular iteration.
Thus, provided that the impedance is not so small that numerical problems occur
when calculating the admittance, and the subsequent factors for the forward and
backward substitution, then convergence of the overall solution between
machines and network will be unaffected. The value of the low-impedance branch
between the dummy and local busbars may be set at a fraction of the total branch
impedance, subject to a minimum value. If this fraction is under 1/100, the change
in branch impedance is very small compared to the accuracy of the network data
input and it is unnecessary to modify the impedance of the branch from the
remote to the dummy busbar.
MACHINE SWITCHING

Machine switching may be considered, either as a network or as a machine

operation. It is a network operation if a dummy busbar is created to which the
machine is connected. The dummy busbar is then connected to the original
machine busbar by a low-impedance branch.
Alternatively, it may be treated as a machine operation by retaining the
original network topology. When a machine is switched out, it is necessary to
remove its injected current from the network solution. Also, any shunt
admittance included in the network Y matrix, which is due to the machine, must
be removed.
Although a disconnected machine can play no direct part in system stability, its
response should still be calculated as before, with the machine stator current set
to zero. Thus machine speed, terminal voltage, etc., can be observed even when
disconnected from the system and in the event of reconnection, sensible results
are obtained.
Where an industrial system is being studied many machines may be
disconnected and reconnected at different times as the voltage level changes. This
process will require many recalculations of the factors involved in the forward
and backward substitution solution method of the network. However, these can
be avoided by using the method adopted earlier to account for synchronous
machine saliency. That is, an appropriate current is injected at the relevant
busbar, which cancels out the effect of the shunt admittance.

9.7. INTEGRATION
Many integration methods have been applied to the power-system transient
stability problem and the principal methods are discussed in Appendix I. Of
these, only three are considered in this section. They are simple and easily applied
methods which have gained wide acceptance. The purpose of the third method is
not to provide another alternative but to clarify the differences between the other
two methods.
Explicit Runge-Kutta methods have been used extensively in transient
stability studies. They have the advantage that a'packaged' integration method is
usually available or quite readily constructed and the differential equations are
incorporated with the method explicitly. It has only been with the introduction of
more detailed system component models with very small time constants that the
problems of stability have caused interest in other methods.
Fourth-order methods (p = 4) have probably been the most popular and
among these the Runge-Kutta Gill method has the advantage that round-off
error is minimized. With reference to equations (1.4.1) to (1.4.3), for this method
the number of function substitutions is four ( v = 4) and:
 The characteristic root of this fourth-order method, when applied to equation
(1.3.3), is:

and to ensure stability, the step length h must be sufficiently small that z, is less
than unity.
The basic trapezoidal method is very well known, having been established as a
useful method of integration before digital computers made hand calculation
redundant.
More recently an implicit trapezoidal integration method has been developed
for solving the multi-machine transient-stability problem(10),and has gained
recognition as being very powerful, having great advantages over the more
traditional methods.
The method is derived from the general multistep equation given by equation
(1.3.2) with k equal to unity and is thus a single-step method. The solution at the
+
end of n 1 steps is given by:

It has second-order accuracy with the major term in the truncation error
being -&h3.
The characteristic root when applied to equation (1.3.3) is:
Z, = 1 -2bn+,
where

If Re(R)<O then 0 5 bn+,I 1.0 and lz, 1 I 1.0. The trapezoidal method is
therefore A-stable, a property which is shown in Appendix I to be more important
in the solution process than accuracy. The trapezoidal method is linear and thus
in a multivariable problem, like power-system stability, the method is Z-stable.
It can be shown that an A-stable linear multistep method cannot have an order
of accuracy greater than two, and that the smallest truncation error is achieved by
the trapezoidal method. The trapezoidal method is thus the most accurate 1-
stable finite difference method possible.
The method, as expressed by equation (9.7.4), is implicit and requires an
iterative solution. However, the solution can be made direct by incorporating the
differential equations into equation (9.7.4). Rearranging forms algebraic equa-
tions as described in Appendix I.

Fig. 9.16. Simple transfer function

For example, consider the trivial transfer function shown in Fig. 9.16. The
differential equation for this system is given by:

with the input variable being denoted by 'z' to indicate that it may be either
integrable or nonintegrable.
The algebraic form of equation (9.7.7) has a solution at the end of the (n + 1)th
step of:

where

and

Provided that the step length h remains constant it is unnecessary to re-

evaluate b or m at each step. That is:

There is little to be gained by this, however, as it is a simple process and it is often

desirable to change h during a study.
A comparison between the Runge-Kutta Gill and the trapezoidal methods
when used to solve two power system transient stability problems is given in
Tables 9.2 and 9.3. The comparison is made in terms of maximum error (based on
results using very small step lengths) and central processor unit (CPU) execution
time.
The advantages of the C-stable trapezoidal method are apparent from both
tables, but the results are sufficiently different to show that an absolute
comparison between methods cannot be made. The nonlinearity of the equations
 233
in any system also effect the errors obtained. CPU time using the Runge-Kutta
Gill method is a function of the step length but this is not so with the trapezoidal
method. For very small step lengths, only one iteration per step is needed using
the trapezoidal method but as the step length increases so does the number of
iterations. The relationship between step length and iterations is nonlinear, with
the result that there is an optimum step length in which the iterations per step are
small but greater than one.
For comparison, the backward Euler method is also included. This is a first-
order method with the solution given by:

Table 9.2

Runge-Kutta Gill Trapezoidal Backward Euler

Step length Max. error CPU time Max. error CPU time Max. error CPU time
(m4 (degs) (4 (degs) (4 (deJ3s) 6)

Table 9.3

Runge-kutta Gill Trapezoidal Backward Euler

Step length Max. error CPU time Max. error CPU time -Max. error CPU time
( 4 (degs) (4 Wgs) (4 (degs) (4
10.0 8.6 1.67 0.5 2.37 - -
5.0 4.4 3.06 0.1 2.3 1 8.5 2.76
2.0 1.7 7.24 - 3.74 3.8 3.64
1.O 1.2 14.19 - 7.12 1.8 6.80
0.5 0.9 28.00 - 13.88 0.6 13.24

and the characteristic root when applied to equation (1.3.3) is:

Despite the three orders of accuracy difference between it and the Runge-Kutta
Gill, the backward Euler method compares well.
The results for the trapezoidal and backward Euler methods were obtained
using linear extrapolation of the nonintegrable variables at the beginning of each
step. This required the storing of machine terminal voltages and currents together
with other nonintegrable variables obtained at the end of the previous step.

Problems with the trapezoidal method

Although the trapezoidal method is C-stable and the step length is not
constrained by the largest negative eigenvalue, the accuracy of the solution
corresponding to the largest negative eigenvalues will be poor if a reasonable step
length is not chosen.
With the backward Euler method, the larger the step length the smaller the
characteristic root

whereas for the trapezoidal method

For small step lengths the characteristic roots of both methods tend towards, but
never exceed, unity (positive).
ZlcBE) and z,(,,,~) + +1 as h l + 0
The effect of too large a step length can be shown in a trivial but extreme
example. The system in Fig. 9.16 and equation (9.7.7) with a zero time constant
T, and unity gain G, is such an example.
If the input z(t) is a unit step function from an initial value of zero, then with a
zero-time constant, the output y(t) should follow the input exactly, that is a
constant output of unity. In fact, the output oscillates with y, = 2, y, = 0, y, = 2,
etc.
Table 9.4 shows the effectof different step lengths on this simple system with a
nonzero-time constant T. This table shows that oscillations occur when h l is
smaller than - 2, that is, when the characteristic root z, is negative. The
oscillations decay with a rate dependent on hl, that is, the rate is dependent on the

Table 9.4. The effect of different step lengths on the solution of a simple system (Fig. 9.16) by the
trapezoidal method

Step Trap Exact Trap Exact Trap Exact Trap Exact

No. method soh. method soln. method soln. method soh.
magnitude of 2,. It can also be seen that accuracy is good provided that hA is
greater than or equal to - 0.5.
Oscillations are only initiated at a discontinuity. Provided that there is no step
function input, the output of a transfer function with zero-time constant
duplicates the input.
The example given is an extreme case and for the power-system stability
problem this usually only occurs in the input circuit of the AVR.
For the mechanical equation of the synchronous machine, the speed is given
by:

where Pa is the accelerating power given by Pa = P, - Pe, and the damping

factor Da is zero. Therefore, in this case

where

and oscillations do not occur. Da, when it does exist, is usually very small and any
oscillations will similarly be very small.
For the electrical equations of the synchronous machine, only the current can
change instantaneously, and the effect is not as pronounced as for a unit step
function.
Techniques'' 2, are available to remove the oscillations but they require a lot of
storage and it is simpler to reduce the step length.

Programming the trapezoidal method

There is no means of estimating the value of errors in the trapezoidal method but
the number of iterations required to converge at each step may be used as a very
good indication of the errors. As previously mentioned, the number of iterations
increases more rapidly than the step length and thus the number of iterations is a
good reference for the control of the step length. It is suggested(13)to double the
step length if the number of iterations per step is less than 3 and to halve it if the
number of iterations per step exceed 12. The resulting bandwidth (3 to 12) is
necessary to stop constant changes in the step length.
To avoid problems of step length chattering a factor of about 1.5 (instead of 2)
may be used. Unfortunately it is difficult to maintain a regular printout interval if
a noninteger factor is used.
Even using a step length changing factor of 2, it is difficult to maintain a regular
printout interval. Step halving can be carried out at any time but indiscriminate
step doubling may mean that there is no solution at the desired printout time.
Doubling the step length thus should only be done immediately after a printout
and the step length should not be allowed to exceed the printout interval.
On rare occasions, it is possible that the number of iterations at a particular
step greatly exceeds the upper desired limit. It can be shown that the convergence
pattern is geometric and usually oscillatory ( I 3 ) after the first five or six iterations.
Even when diverging, the geometric and oscillatory pattern can be observed.
Schemes can thus be devised which estimate the correct solution. However, these
schemes are relatively costly to implement in terms of programming, storage and
execution and a more practical method is to stop iterating after a fixed number of
iterations and start again with a half-step. It is not necessary to store all the
information obtained at the end of the previous step, in anticipation of a restart,
as this information is already available for the nonintegrable variables if an
extrapolation method is being used at the beginning of each step. Further, much
information is available in the C and M constants of the algebraic form of the
integration method.
For example, with the two-variable problem given by equation (9.7.7),if z(t)is a
nonintegrable variable, then its value at the end of the nth step znis stored. The
value of the integrable variable y(t) at the end of the (n)th step yn can be re-
evaluated from equations (9.7.9) and (9.7.10) to be:

In only a few cases where the differential equation is complex need the value of yn
be stored at the beginning of each step. While the method requires programming
effort it is very economical on storage and the few instances where it is used do not
affect the overall execution time appreciably.
Linear extrapolation of nonintegrable variables at the beginning of each step is
a very worthwhile addition to the trapezoidal method. Although not essential, the
number of iterations per step is reduced and the storage is not prohibitive. Higher
orders of extrapolation give very little extra improvement and as they are not
effective until some steps after a discontinuity their value is further reduced.
It is only at the first step after a discontinuity that linear extrapolation can not
be used. As this often coincides with a large rate of change of integrable variables,
the number of iterations to convergence can be excessive. This is overcome by
automatic step length reduction after a discontinuity. Two half-step lengths,
before returning to the normal step length, has been found to be satisfactory in
almost all cases.(13)

Application of the trapezoidal method

The differential equations developed in this chapter have all been associated with
the synchronous machine and its controllers. These equations can be transfor-
med into the algebraic form of the trapezoidal method given by equation (9.7.8).
While these algebraic equations can be combined to make a matrix equation this
has little merit and makes discontinuities such as regulator limits more difficult to
apply
In order to simplify the following equations, the subscripts on the variables
have been removed. It is rarely necessary to retain old values of variables and,
where this is necessary, it is noted. The variable values are thus overwritten by
new information as soon as they are available. The constants C and M associated
with the algebraic form are evaluated at the beginning of a new integration step
and hence use the information obtained at the end of the previous step.
SYNCHRONOUS MACHlNE

The two mechanical differential equations are given by equations (9.2.5) and
(9.2.6) and the algebraic form is:

where
C , = (1 - 2 . M;Da)o + M,(Pm - Pe + 4nlfo. Da)

and also

where

It would be possible to combine these equations to form a single simultaneous

solution of the form
6 = C i + ML(Pm - Pe) (9.7.25)
where

but machine speed o is a useful piece of information and would still require
evaluation in most problems.
It is also more convenient to retain the electrical power (Pe)as a variable rather
than attempt to reduce it to its constituent parts:

Pe = I,. Vd + I; Vq+ (I: + I i ) R a (9.7.26)

Thus Pe is extrapolated after C , and M , have been evaluated.
The mechanical power P m is an integrable variable which, in the absence of a
speed governor model for the machine, is constant.
There are four electrical equations associated with the change in flux in the
synchronous machine and these are given by equations (9.2.13), (9.2.14), (9.2.17)
and (9.2.18). The algebraic form of these equations are:
E; = C, + M,(Ef + ( X , - X;)Id) (9.7.27)
where
C, = (1 - 2Mq)E; M,(Ef + + ( X , - Xi)I,)
M, = h/(2Ti0 + h)
also

where
Cd = (1 - 2Md)E;- M d ( X q- X;)I,
M d = h/(2Ti0 h) +
also
Ei = C,, + M,,[Ei + ( X ; - XC)ld)
where
C,, = (1 - 2Mqq)E: M,,(Ei+ +( X i-Xi)ld)
M,, = h/(2Tio + h)
and also
E; = Cdd + Md(E;- ( X i - X p , )
where
cdd= (1 - 2Mdd)Ef;+ Mdd(E;- ( X i - X;)I,)
M,, = h/(2Tio + h)
SYNCHRONOUS MACHINE CONTROLLER LIMITS
There are usually limits associated with AVRs and speed governors and these
require special consideration when applying the algebraic form of the trapezoidal
rule. It is best to ignore the limits at first and develop the whole set of 'limitless'
equations. Rather than confuse this discussion, it is easier to consider a simple
AVR system as shown in Fig. 9.17, for which can be written:

Fig. 9.17. Block diagram of a simple controller

 "in

Fig. 9.18. Modified block diagram of a simple controller of

Fig. 9.17

subject to

The feedback loop can be rearranged to avoid the derivative of Vout being
explicitly required as described in Section 9.3 and this is shown in Fig. 9.18.
Equation (9.7.32) is now replaced by:

Equations (9.7.31) and (9.7.34) can be transformed into the algebraic form:

where C,, C,, M 1 and M, may be determined in the usual way.

A simultaneous solution for the whole system is now possible by combining
equations (9.7.33), (9.7.35) and (9.7.36) to give:

and

After a solution of Voutis obtained it may be subjected to the limits of equation

(9.7.31). If it is necessary Vf b can now be evaluated from equations (9.7.33) and
(9.7.36) using the limited value of V,,/,,,.
Where this simple controller model represents an AVR the input Vinmay well
be the (negated) deviation of terminal voltage Vt from its specified value (K). It is
simpler to treat Vt as an extra nonintegrable variable rather incorporate
V, = J(V; + v:)
in the model.
The usual models of speed governors do not have feedback loops associated
with them, but the input to the governor (machine speed) is related to the turbine
output (mechanical power) by differential equations. It is therefore necessary to
solve a set of simultaneous equations in a similar manner to the example above.
The simultaneous solution should be first made at a point at which limits are
applied (i.e. at the valve) and then, after ensuring the result conforms to the limits,
all the other variables around the loop (including machine speed and rotor angle)
can be evaluated.

SOLUTION FOR SATURATING AVR EXCITER

Another problem occurs when a nonlinear function is encountered. Equations
(9.3.3)and (9.3.1 1)may be combined to form a single differential equation but this,
then, involves a term in Ef which complicates the evaluation of Ef. As the
saturation function is approximate, it may be further simplified to give:

where Ef* is the value of Ef at the previous iteration and k: and kz are
determined from Ef *. The equation describing the saturating exciter is thus:

and applying the trapezoidal rule the algebraic form of solution is:
+
Ef = LCe,- Me,-(Vam)]/[l + Me,-(k: Ef * - kz - Ke,-)] (9.7.40)
where
+ Ke,-)Me,-)Ef + Me,-Vam
Ce,- = (1 - 2(Ke
Me,-= hl(2Te + h(Ke + Ke,-))
Ke,-=k,Ef-k2
Cef, Me,- and Ke,-are evaluated once at the beginning of the step and, hence, only
contain information obtained at the end of the previous step.

9.8. STRUCTURE OF A TRANSIENT STABILITY PROGRAM

Overall structure
An overview of the structure of a transient stability program is given in Fig. 9.19.
Only the main parts of the program have been included, and as can be seen, the
Section 1

I I
1 Read in steady-statesystem data

Read in switching data

I

2 . Calculate sparse factored inverse of

unfaulted nodal network matrix
(bifactorization 1

Section 3 I 2. Make switching chonges ( i f necessary 1

Section 4

Section 5
I
r
1
1
Etther Solve for machinesand network iteratively
or recaIculatenoninte~rabIevariables
(initially and when iwftching has occurred 1
I

2. Make power balance check of initial

YES
I
1

Store initial Reset initial

Fig. 9.19. Overview of a transient stability program structure

same system may have several case studies performed on it by repeatedly
specifying switching data. When no further switching data is available, control
returns to the start to see if another system is to be studied.
With care, the program can be divided into packages of subroutines each
concerned with only one aspect of the system.(13)This permits the removal of
component models when n i t required and the easy addition of new models
whenever necessary. Thus for example, the subroutines associated with the
synchronous machine, the AVRs, speed governors, etc., can be segregated from
the network. Figure 9.20 shows a more detailed block diagram of the overall
structure where this segregation is indicated. The diagram is subdivided into the
five sections indicated in Fig. 9.19.
While the block diagrams are intended to be self-evident several logic codes
need to be explained. These are:
KASE-This is the case study number for: a particular system. It is initially set
to zero and incremented by 1 at the end of the initialization and at the end of
each case study.
KBIFA1-The sparse factored inverse of the nodal network matrix is obtained
using three bifktorization subroutines. The first and second subroutines are
integer routines which determine busbar ordering and nonzero element
location. The code KBIFAl is set to unity if it is necessary to enter these two
subroutines, otherwise it is set to zero.

Network calculations Synchronous AVR calculations Speed governor

machine calculations
calculations
-
--c
___
*-_
___
.(

KBI FA1 = 0
KBIFA3.0

Network AVR data

data input data input input

I @%+@ data?

( a 1 Section 1
 Switching
data input

order and nonzero

element location

and Is

initial conditions
conditions

YES initial
conditions

Perform
numerical part
of, bifactorization

Print out initial

conditions

( b ) Section 2
 switching ops.
if branch change
set KBIFA3 = 1
if new branch

order and non zero

element location

part of

( c ) Section 3
 1 For further

r Print-out
busbar and
branch results
if required
Print-out
sync. machine
results

i
AVR results
if reouired

Print-out
speed gov.
results if
required

7
11
Set flag for step

( d ) Section 4
 TIME <MAXTI >
1 Perform oower
balonc; check
to confirm
initial data
1
I

d
,
i End of case'

* KASE =I?

( steady-state
conditions
1

Reset initiol
Isteady-state )
conditions

(e Section 5
Fig. 9.20. Structureof transient stability program (a) Section 1 (b) Section 2
(c) Section 3 (d) Section 4 (e) Section 5

KBIFA3-The elements of the sparse-factored inverse are evaluated in the

third bifactorization subroutine. The code KBIFA3 is set to unity if it is
necessary to enter this subroutine, otherwise it is set to zero. When KBIFA3
is unity, it indicates that a network discontinuity has occurred and hence it is
also used for this purpose.
TIME-The integration time.
H-The integration step length. Like KBIFA3, it is also used to indicate a
discontihuity when it is set to zero.
PRINTTIME-The integration time at which the next print out of results is
required.
MAXTIME-The predefined maximum integration time for the case study.
ITMAX-Maximum number of iterations per step since last print-out of
results.
 Note that many data error checks are required in a program of this type but
they have been omitted from the block diagram for clarity.

Structure of machine and network iterative solution

The structure of this part of the program requires further description. Two forms
of solution are possible depending on whether an integration step is being
evaluated or the nonintegrable variables are being recalculated after a discon-
tinuity. A block diagram is given in Fig. 9.21.
The additional logic codes used in this part of the program are:
ERROR-The maximum difference between any integrable variable from one
iteration to another.
ITR-Number of iterations required for solution.
IHALF-Number of immediate step halving required for the solution.
TOLERANCE-Specified maximum value of ERROR for convergence.

in case H is reduced

C and M for

nonintegrable ---------+c!-~~zi%tGt1
dusually required
L------.J
I

II alaebraic form of
tr~pezoidolmethod
-Same
-Lspeed
for each
gov.
1
I

ITR = 0
IHALF = 0

(a Section 1
 ITR = ITR +1

NO

Same for each Calculate currents

nonlinear + injected into network
load for each sync.
machine
J.
Calculate bus
voltages from
bifactored
network

machine
non -integrable ---------~5;Gd;ov_colGot;
--------I usually required I
L ------- J

Evaluate integrable
variables using Same for each A V ~
algebraic form of =I Same for each 1
(
trapezoidal method
also determine
ERROR I+ speed gov.
1

Y
between iterations

YES

1 Set flag for step

( b 1 Section 2
 Re-evaluate
conditions
at beginmg of

*
T
0
Not converging'

TIME = MAXTIME

Exit

(c Section 3

Fig. 9.21. Structure of machine and network iterative solution (a) Section l
(b) Section 2 (c) Section 3

If convergence has not been achieved after a specified number of iterations, the
case study is terminated. This is done by setting the integration time equal to the
maximum integration time. The latest results are thus printed out and a new case
study is attempted.

9.9. GENERAL CONCLUSIONS

The transient stability program described in this chapter is sufficient for many
basic stability studies. It is more than adequate when first swing stability is being
evaluated and the machine detail and controllers will allow second and
subsequent swing stability to be examined also.
However, if synchronous machine saturation or compound thermal turbines
have to be modelled, it will be necessary to incorporate parts of Chapter 10 into
the program. The structure set out at the end of this chapter should allow changes
of this sort to be made quite easily. Similarly,if other system components are to be
included this can be done without difficulty.

9.10. REFERENCES
1. 0. I. Elgerd, 1971. Electrical Energy Systems 7heory: An Introduction. McGraw-Hill,
New York.
2. B. M. Weedy, 1979. Electric Power Systems. Wiley and Sons, London.
3. C. Concordia, 1951. Synchronous Machines. 7heory and Performance. Wiley and Sons,
New York.
4. E. W. Kimbark, 1956. Power System Stability :Synchronous Machines. (Vol. 3). Wiley
and Sons, New York.
5. P. L. Dandeno, et al. 1973. 'Effects of synchronous machine modeling in large-scale
system studies', I E E E Transactions on Power Apparatus and Systems, PAS92 (2), 574-
582.
6. IEEE Committee Report, 1968. 'Computer representation of exciter systems', IEEE
Transactions on Power Apparatus and Systems, PAS87 (6), 1460-1464.
7. IEEE Committee Report, 1973. 'Dynamic models for steam and hydro turbines in
power-system studies', IEEE Transactions on Power Apparatus and Systems, PAS-92
(6), 1904-1 915.
8. P. L. Dandeno and P. Kundur, 1973. 'A noniterative transient stability program
including the effects of variable load-voltage characteristics', l E E E Transactions on
Power Apparatus and Systems, PAS92 (9, 1478-1484.
9. G. L. Berg, 1973. 'Power system load representation', Proc. I E E 120 (3), 344-348.
10. H. W. Dommell and N. Sato, 1972. 'Fast transient stability solutions', IEEE
Transactions on Power Apparatus and Systems, PAS-91 (4), 1643-1650.
11. A. Brameller, et al. 1969. Practical Diakoptics for Electrical Networks. Chapman and
Hall, London.
12. L. Lapidus and J. H. Seinfeld, 1971. Numerical Solution of Ordinary Differential
Equations. Academic Press, New York.
13. C. P. Arnold, 1976. 'Solutions of the multi-machine power-system stability problem.
Ph.D. thesis, Victoria University of Manchester, U.K.
 Power System Stability-
Advanced Component
Modelling
10.1. INTRODUCTION
This chapter develops further some of the component models described in
Chapter 9 and introduces new models needed to investigate the effects of other
a.c. system plant components. Turbine generator models are extended by
considering the effects of saturation in the synchronous machine and the response
of compound thermal turbines. Detailed consideration is also given to the
modelling of induction motors and static power converters. The chapter also
deals with protective gear modelling and unbalanced faults.
The induction motor model allows for a good representation over the whole
speed range so that motor starting can be investigated. The model can be created
in three ways depending upon the induction motor data available.
Basic formulation of three-phase bridge rectification and inversion has already
been considered in Chapter 3 and here it is extended so that the dynamic model
can include abnormal operating conditions encountered during stability studies.
It must be clarified however, that the controllability of h.v.d.c. links during large
disturbances in either the a.c. or d.c. system cannot be determined by the type of
transient stability program described in this chapter. These and other
problems associated with transient stability analysis involving h.v.d.c. links are
considered in Chapter 13.
The grouping of subroutines relevant to a particular component of the power
system or aspect of the study, as developed in Chapter 9, should be retained for
the models produced in this chapter. This ensures that additional models can be
incorporated easily and models removed when not necessary.

10.2. SYNCHRONOUS MACHINE SATURATION

The relationship between mutual flux and the exciting MMF within a machine is
not linear and some means of representing this nonlinearity is necessary if the
results obtained from a stability study are to be accurate.
 In most multimachine stability programs, each machine is represented by a
voltage behind an impedance. As explained in Chapter 9 the impedance consists
of armature resistance plus either transient or subtransient reactance. Also the
voltage magnitude may be fixed or time-varying, depending on the complexity of
the model.
Saturation may thus be taken into account by modifying the value of the
reactance used in representing the machines. However, as explained for the model
developed in Chapter 9, it is more convenient to fix the reactance and adjust the
voltage accordingly.
Saturation is a part of synchronous machine modelling where there is still
uncertainty as to the best method of simulation. The degree of saturation is not
the same throughout the machine because the flux varies by the amount of
leakage flux. Also, the saturation in the direct and quadrature axes are different,
although this difference is small in the case of a cylindrical rotor.
Various methods have been adopted to account for saturation which differ not
only in the model modification technique but also in the representation of the
saturation characteristic of the machine.

Classical saturation model

Classical theory") for a cylindrical rotor machine assumes that the saturation is
due to the total MMF produced in the iron and is the same in each axis.
It is necessary to make further assumptions in order to simplify the model as
follows:
(1) The magnetic reluctance in both axes are equal. Thus, the synchronous
reactances are equal, i.e. X , = X,.
(2) Saturation does not distort the sinusoidal variations assumed for rotor and
stator inductances.
(3) Because load-test data is not usually available, saturation is determined
using the open-circuit saturation curve.
(4) Potier reactance X , may be used in calculating saturation.
(5) The total iron MMF (Fe) may be determined from

where S is the saturation factor, defined as:

iron MMF
S=l+
air gap MMF
(6) With reference to Fig. 10.1, the saturation factor S may be determined
from :

Figure 10.2 shows a typical voltage and MMF diagram for a round rotor
synchronous machine. Potier voltage Ep, the voltage behind Potier reactance,
 characteristic

Saturation factor ( S ) voltage V

=AC
A6

Field current (MMF)

Fig. 10.1. Open circuit saturationcharacteristic of a synch-

ronous machine

Ff

Fig. 10.2. Vector diagram of cylindrical rotor synch-

ronous machine saturation
may be determined readily from the terminal voltage Vt and the terminal current
I. The MMF required to produce this voltage is found from the open circuit
saturation curve of Fig. 10.1. Armature reaction F is found using Assumption 4
from which the field MMF (Ff) is calculated. The voltage equivalent to Ff
referred to the stator is Ef. It is readily apparent that rotating the MMF diagram
through 90" gives:
Ff a Ei (in the steady state)
FeaS.Ep (10.2.4)
F a I(Xd- Xp)
In Fig. 10.2, the reactance X, is the saturated value of X,. This produces an
internal machine voltage E, which lies on the quadrature axis. As I(X, - Xp) is
parallel with I-X,,, then
EP -I(X,-Xp)
-- -E,
-
S. Ep I(X, - Xp) Ei
and from this
Ei = S. Eq
and

All machine reactances subject to saturation are similarly modified.

Salient machine saturation

In the case of a salient synchronous machine, it may be assumed that the direct
and quadrature axis armature reaction MMF's (F, and F, respectively) are
proportional to the reactive voltage drops I,.Xa, and I,.Xa, respectively.
Assumptions 3 and 4 for the classical model also apply.
There are many different methods of accounting for the saturation effect. The
methods considered here assume that saturation in the d-axis is due at least in
part to the component of flux in the d-axis. The first method ignores saturation in
the q-axis, the second method accounts for quadrature axis saturation by the
component of flux in the q-axis, the third method considers that the total flux
contributes to the saturation in both axes.
In the first method, the density of the flux due to the quadrature axis armature
reaction MMF (F,) is considered sufficiently small that saturation effects on
voltages are thus neglected in the direct axis. The other component of the
armature reaction MMF (F,) adds directly to the field MMF (Ff ) to produce a
main flux which in turn produces a quadrature axis voltage subject to saturation.
The saturation level is determined by the quadrature component of Potier
Voltage (Ep,).
The second method,(2)allows for saturation in both the direct and quadrature
axes components of the Potier voltage. It is assumed that the reluctances of the
d-axis and q-axis paths differ only because of the different air gaps in each axis. The
d-axis component of Potier Voltage (Ep,) is thus modified by the ratio X,/X,
before the q-axis saturation factor is determined. Provided that it is assumed that
the vector sum of the two saturated main flux components (Fe, and Fed) is in
phase with the MMF proportional to Potier Voltage, then the saturated d and q
axis synchronous reactances (X, and Xqs)are:

A third method'') distinguishes between the saturation in the rotor and stator,
and saturation factors based on Ep and Ep, are obtained. This method is dficult
to implement because it is necessary to ensure that saturation is not applied twice
to any part of the machine. That is, the saturation in the field poles must be
isolated from that of the armature, giving two saturation curves. The two
saturation factors for this case may be defined as:
iron MMF in the stator
S,=1+
total air gap MMF
iron MMF in the rotor
S,=1+ (10.2.11)
direct axis air gap MMF
where S,, acts equally on both d and q axes and S, acts on the direct axis only.
Figure 10.3, 10.4 and 10.5 demonstrate the differences between the three
methods of saturation representation.

Fig. 10.3. Salient pole synchronous machine with direct axis saturation only
 Fig. 10.4. Salient pole synchronous machine with direct and quadrature axis saturation

Fig. 10.5. Salient pole synchronous machine with separate stator and rotor saturation

Simple saturation representation

An even simpler method of including the effect of saturation is to calculate the
saturation initially (by some means) after which it is either held constant, or
varied according to the slope of the saturation curve at the initial point. This
method is suitable for small perturbation studies where Potier voltage and
machine angle do not vary greatly.
Saturation curve representation
The open-circuit saturation curve must be stored within the computer so that a
new saturation factor can be determined at every stage of the study.
The most accurate method of storing this curve is to fit a polynomial of the
form :
If=C,+C,V+C2V2+C3V 3 . . . + c n v n (10.2.12)

by taking n + 1 points on the curve. Normally n would be 5, 7 or 9. This is a

clumsy method of both entering the data and storing it. In multimachine
transient stability studies, where the machines are represented at best by
subtransient parameters and an approximation to Potier reactance is made,
nothing is achieved by such an elaborate method of representing the saturation
curve.
The problem can be simplified by assuming that most of the coefficients of the
polynomial are zero. A sufficiently good approximation is achieved with the
equatiod3):

where n is normally either 7 or 9. Only one point is needed to specify the curve and
if If is always specified at a predetermined voltage, the data entries required per
curve are reduced to one, from which Cn may be readily determined.

Potier reactance
The Potier reactance of a machine is rarely quoted, although the open-circuit
saturation curve is normally available. In order to model the saturation effects, it
is thus necessary to estimate this reactance.
From knowledge of the leakage reactance XI, Be~kwith'~) calculated that:
Xp = XI + 0.63(X; - XI) (10.2.14)

and if X1 is not available, then:

Equation (10.2.15) may be modified to account for the type of synchronous

machine,") i.e.

for a salient-pole machine, as most of the saturation occurs in the poles, and
Xp = 0.7X; (10.2.17)
for a round rotor machine, as most of the saturation occurs in the rotor teeth and
the Potier and leakage reactances have similar values.
The effect of saturation on the synchronous machine model
Saturation effectively modifies the ordinary differential equations describing the
behaviour of the voltages used to model the synchronous machine. Equations
(9.2.13),(9.2.14),(9.2.17)and (9.2.18) become respectively:
+
PEI,= ( E f ( X d- X;)Id - SdEI,)/Ti0 (10.2.18)
PEA = ( - ( X , - XI,)I, - SqE;)/TI,o (10.2.19)
~ (SdEi
P E= + ( X , - X,")Id- S,E;)/T,", (10.2.20)
~ (S,E; - (XI, - X;)I, - SqE&')/Ti0
P E= ( 1 0.2.21 )
where S, and S, are the direct and quadrature axes saturation factors.
Where subtransients are considered then equations (9.2.15) and (9.2.16) are
replaced by :
Ei - Vt, = Ra-I, - Xis.Id (10.2.22)
Ei- Vt,=Ra.I,+XC.I, (10.2.23)

Representation of saturated synchronous machines in the network

Representation of a salient but unsaturated synchronous machine in the network
has been discussed in Section 9.6. When saturation occurs, a double adjustment
must be made at each step in the solution process.(6)
With the notation developed in Section 9.6, the fixed admittance (Yo)which is
included with the network is made up from unsaturated and nonsalient values of
reactance, whereas the correct admittance (Y,,) is made up from saturated and
salient values. Using Fig. 9.14, as before the adjusting current to account for this
change in admittance is

That is :

Ra Xdqs 1
-
(Ra2+ X,"X;)
- xaS fi
The current I!$, is similar to iadj
developed in Section 9.6:

The current injected into the network is given by equation (9.6.4) and as the
terms in the brackets contain no saliency then in the real and imaginary axis of the
network:

where i!& contains saturated but nonsalient reactance terms and I$:, contains
salient and saturated reactance terms.

Note that the third part of equation (10.2.27) is yo Vand not yo E". This part of the
injected current is merely the current flowing through yo and could be eliminated
if Yowas not included in the network. The conditioning of the network would be
reduced however, and in certain systems this could lead to numerical problems.

Inclusion of synchronous machine saturation in the

transient stability program
Only two subroutines need modification to allow saturation effects in synch-
ronous machines to be modelled. In both cases, an iterative solution is necessary
for each saturating machine, although in most instances the number of iterations
is small.
Saturation is a function of the voltage behind armature resistance and Potier
reactance. Assuming the second method of salient machine saturation is being
used, then from equation (10.2.13):

where

and

and I d and I, are given by equations (10.2.22) and (10.2.23).

A Jacobi iterative technique is quite adequate to establish the initial conditions
of the synchronous machine and this can be incorporated in the relevant
subroutine shown in Section 2 of Fig. 9.20.
During the time solution, however, saturation can vary over a large range of
values and a Newton form of iteration is an advantage especially if large
integration steps are used.
Redefining equation (10.2.30) as:

the elements of a 2 x 2 Jacobian matrix can be found. However, elements f l/aS, a

and a f 2/aSd are small with respect to the other two elements and jf Ra is
considered to be zero then the four elements reduce to:

af i -[(n- ~ ) . C , , ( E ~ , ) " - ~ - X ~-
-- ( XXp)(E;
; - Vt,)]
-1
as, [(sd - l l X p + X&'l
a
- -f 2 - [ ( n - l ) . C n q ( E p d ) n - 2 . X p (-
~ ;Xp)(E," - V t , ) ]
-1 (10.2.34)
as, C(S, - ~ ) X + P xi1

This decouples the Newton method and each saturation factor may be solved
independentl~'~)
sy + 1) , -f 1(p)I(af l I a s d ) c p )
S(P)
d (10.2.35)
sr ) = S:) - f ~ l ( a2/asqyr)
+ f
Despite the advantages of a Newton form of solution, it can be found to be
divergent if too great an integration step length is used. Analysis of the functions
f 1 andf 2 show that they have discontinuities, when XC = (S, - 1)Xp and Xi =
(S, - 1)Xp respectively, although otherwise are almost linear. It is therefore
necessary to monitor this iterative procedure and modify the step length if
necessary.
The evaluation of S, and S, should be performed twice during each iteration.
Considering Fig. 9.21, this is during the calculation of the injected currents into
the network and the calculation of the nonintegrable variables. Provided the
discontinuity is not encountered, convergence is achieved in one or two iterations
at each re-evaluation especially if the saturation factors are extrapolated at
the beginning of each step.

10.3. DETAILED TURBINE MODEL

More detailed turbine models than the one described in the previous chapter are
required for the following reasons:
A longer-term transient stability study or a dynamic stability study is to be
made.
The turbine is a two-shafted cross-compound machine which has a
separate generator on each shaft.
Generator overspeed is such that an interceptor valve may operate during
the study.
A generalized model to accommodate the different types of compound turbine
has been developed by the IEEE.'') As with the generalized AVR model, by
setting certain gains to either zero or unity and time constants to either infinity
(very large) or zero, the model can be reduced to any desired form. An interceptor
valve can easily be incorporated as shown in Fig. 10.6.

Fig. 10.6. Generalized detailed turbine model including H.P. and interceptor valves

Some normal compound turbine configurations are shown in Fig. 10.7 and
Table 10.1 gives typical values for these configurations using the generalized
model. A hydroturbine can also be represented and the values given in Table 10.1
 -- Shaft
To condenser

Control
,
Valve valves
position :eS;" Shaft

TO condenser
(b)

[~eheater1 lReheater l 1 Crossover

--- Shaft
1 1
To condenser
(c 1

valve ,%
!:\:
position steam
chest

1Reheater
valve
position
w$kz!
steam HFf IP shaft
chest
1~rossoverj
I

I%--
TO condenser
L P shaft

Control
Valve valves,
position * steam
chest
I Reheater1 I Reheated ICrossover

Fig. 10.7. Common steam turbine configuration^(^) (@ 1982 IEEE)

(a) Nonreheat; (d) Cross compound, single reheat ;
(b) Tandem compound, single reheat; (e) Cross compound, single reheat;
(c) Tandem compound, double reheat; (f) Cross compound, double reheat
Ppv

Fig 10.7. (continued). Approximate linear models(8)(01982 IEEE)

 Table 10.1. Parameters used in generalized detailed turbine model.@)( @ 1982 IEEE)

Time constants Fractions

with typical values (seconds) with typical values (p.u.)

Turbine system Figure T4 Ts T6 T7 K1 K2 K3 K4 K5 K6 K7 K8

Tandemcompound
singlereheat
Tandemcompound
doublereheat
Cross-compound
singlereheat
Cross-compound
single-reheat
Cross-compound
double-reheat
Hydro 9.9a 0 4Tw - - -2 0 3 0 0 0 0 0
are justified by the method of representing a lead-lag circuit described in
Chapter 9, with the time constant T , set at Tw in the case of the simplest model.
The full set of equations for the detailed turbine model is:
pG4 = (P,, - G4)/T4 (10.3.1)
pG5 = (G4 - G s ) / T ~ (10.3.2)
Piv = G5-Pvi (10.3.3)
pG6 = (Piv - G6)/T6 (10.3.4)

P,, =K 1.G4 + K3.Piv + K5.G6 + K7 -G7 (10.3.6)

P,, =K2.G4+ K 4 . P i v + K6.G6+ K8-G7 (10.3.7)
Also note that:

and that, in the initial steady state, the interceptor valve, if present, will be fully
open (Pvi = 1 ) in which case:

The speed governor controlling the interceptor valve is similar to that

controlling the HP turbine except that it is set to operate at some overspeed value
of slip (k,) and not about synchronous speed. Equation (9.3.12)can be modified in
this case to:

10.4. INDUCTION MACHINES

An approach similar to that used to construct the synchronous machine models is
required if induction machines are to be explicitly m ~ d e l l e d . ( ~However,
) * ( ~ ) speed
cannot be assumed to vary only slightly and this basic drfference requires that the
equations describing the behaviour of induction machines be somewhat different
from those developed for a synchronous machine.

Mechanical equations
It is necessary to express the equation of motion of an induction machine in terms
of torque and not power. Also symmetry of the rotor makes its angular position
unimportant, and slip (S) usually replaces angular velocity (a)as the variable,
where:
 Assuming negligible windage and friction losses and smooth mechanical shaft
power, the equation of motion is:
pS = (Tm - Te)/(2Hm) (10.4.2)
where Hm is the inertia constant measured in kWs/kVA established at
synchronous speed. The mechanical torque (Tm) and electrical torque (Te) are
assumed to be positive when the machine is motoring.
The mechanical torque Tm will normally vary with speed, the relationship
depending on the type of load.
A commonly used characteristic is:

where k = 1 for fan-type loads and k = 2 for centrifugal pumps. A more elaborate
torque/speed characteristic can be used for a composite load, i.e.:

which can include the effect of striction when start-up is being considered.
In terms of slip the torque is thus:

where

B a{b + 2c)
C a c
The values of A, B and C are determined from the initial (steady-state)loading of
the motor and hence its initial value of slip.
The electrical torque Te is related to the air gap electrical power by the
electrical frequency which is assumed constant and hence:

Electrical equations
A simplified equivalent circuit for a single-cage induction motor is shown in
Fig. 10.8, with R1 and Xl referring to the stator and R2 and X2 referring to the

Fig. 10.8. Steady-state equivalent circuit of a single cage induction

motor
rotor resistance and reactance respectively. In a similar manner to the transient
model of a synchronous machine, an induction motor may be modelled by a
Thevenin equivalent circuit of a voltage E' behind the stator resistance R1 and a
transient reactance X'. The transient reactance is the apparent reactance when
the rotor is locked stationary and the slip (S) is unity and is given by:
X2-Xm
xt=x1+
( X 2 Xm)+
where X m is the magnetizing reactance of the machine. The rate of change of
transient voltage is given by:
pE' = -j 2 n f S E ' - (E' -j(X.0 - X f ) i l ) / T b (10.4.7)
where the rotor open-circuit time constant T b is:

Tb =
+
( X 2 Xm)
2nf OR2
and the open-circuit reactance X o is:

The reactances are unaffected by rotor position and the model is described in
the real and imaginary components used for the network, that is, in the
synchronously rotating frame of reference. Thus, for a full description of the
model, the following equations are used:

Vm- E:, = R1-11, +X1.Ilr ( 1 0.4.11)

pEi = 2nf OSE; - (E: + (XO- X') I lm)/Tto (10 4.12)
pE:, = - 2nf OSE: - (Ek - (XO- X ' ) I i r ) / T f o (10.4.13)
A transient-stability program incorporating an induction motor model uses
the transient and open-circuit parameters, but it is often convenient to allow the
stator, rotor and magnetizing parameters to be specified and let the program
derive the former parameters.
For completeness, the electrical torque may now be written as:

Electrical equations when the slip is large

Singlecage induction motors have low-starting torques and it is often difficult to
bring them to speed without either reducing the load or inserting external
resistance in the rotor circuit. As a result of the low-starting torque, when the slip
exceeds the point of maximum torque, the single-case model is often insufficiently
accurate. These problems are overcome by the use of a double-cage or deep-bar
rotor model.
CAGE FACTOR

When a torque slip characteristic of the motor is available, then a simple solution
is to modify the torque-slip characteristic of the single-cage motor model.
Double-cage or deep-bar rotors have a resistance and reactance which varies
with slip. A cage factor Kg can be included which allows for the variations of
rotor resistance.

where R2(0) is the rotor resistance at zero slip.

It is usually convenient to make the cage factor larger than that necessary to
describe the change in rotor resistance. In this way, the torque-slip characteristics
of the model can be made similar to that of the motor without the need to vary the
rotor reactance with slip. The result of varying the rotor resistance is to modify
the open-circuit transient time constant only, and this can be done quite simply at
each integration step of the simulation.
Rotor reactance does not vary with slip as greatly as rotor resistance, provided
saturation effects are ignored, and its effect on the open-circuit transient time
constant is thus small. Transient reactance (X') varies with rotor reactance.
However, this variation on the term (X, - X') in equations (10.4.12) and (10.4.13)
is insignificant. Thus, the only major effect of varying rotor reactance is in
equations (10.4.10) and (10.4.11) which requires a technique similar to that
adopted in the synchronous machine model to account for saturation and
saliency. However, the gains obtained in using two-cage factors are insignificant
and a single cage factor varying rotor resistance is usually adopted.
LWUBLE-CAGE ROTOR MODEL

An alternative to the cage factor is the use of a better rotor model, though this is
often restricted by the unavailability of suitable data.
Induction motor loads having double-cage or deep-bar rotors can be
represented in a similar manner to a single cage motor.(10)*(")It is assumed that
the end-ring resistance and that part of the leakage flux which links the two
secondary windings, but not the primary, are neglected. The steady-state

Fig. 10.9. Steady-state equivalent circuit of a double cage induction motor

equivalent circuit shown in Fig. 10.9 can thus be obtained where R 3 and X 3 are
the resistance and reactance of the additional rotor winding. A circuit similar to
that of Fig. 10.8 can be obtained by substituting the two parallel rotor circuit
branches by a single series circuit, where:

At any instant during a transient stability study, the rotor impedance may be
assumed to be the steady-state value given above.
Analysis similar to that used in developing equations (10.4.10) to (10.4.13)
gives :

pE; = - 2nf o - S ( E-
~E 3 + pEi + (Ei - E: - (X' - X")Il,,,)/T;; (10.4.20)
PEL = 2rcf o.S(Ei - Ei') + pEk + (EL - EL + ( X ' - X")I 1,)ITs (10.4.21)

with equations (10.4.12)and (10.4.13)applying also.

The parameters for the model, when the motor has a double-cage rotor are
given by equations (10.4.6),(10.4.8),(10.4.9) and:

T;; =
X3 + ( X 2 .X m ) / ( X 2+ Xm)
2nf OR3
If the rotor is of the deep-bar type, then the parameters of the equivalent
double-cage type may be determined using equation (10.4.16)and (10.4.17).The
rotor parameters at zero slip are:

and at standstill are :

This set of nonlinear equations may be solved using Newtonian techniques but by
substituting:

and
X2-Xx
X3=
(X2 - Xx)
where

---
from assuming R2 $R 3 and X 2 SX 3. --
-
the number of variables reduces to two and a simple iterative procedure yields a
result in only a few iteration^."^) A reasonable starting value is X 2 +x,derived

Representation of induction machines in the network

This is quite simple compared to the representation of a synchronous machine as
neither saliency nor saturation are normally considered in the induction machine
models. They may, therefore, be considered as injected currents in parallel with
fixed admittances.
Modifying equations (10.4.18) and (10.4.19) gives a machine current of:
y(P- EM) (10.4.32)

The injected current into the network which includes P is thus:

H 'inj,

'inj,

where the minus sign confirms the induction machine as being assumed to be
motoring.

Inclusion of induction machine in the transient stability program

This is relatively straightforward using the same format as developed for
synchronous machines. Most induction machines are equipped with contactors
which respond to terminal conditions such as undervoltage and it is sometimes
necessary to model this equipment. The characteristics and logic associated with
contactors are included in Section 10.7 (Relays).

10.5. A.C.-D.C. CONVERSION

The use of high-voltage and/or high-current direct current systems is now
sufficiently widespread to require the inclusion of d.c. converter models as a
standard part of a comprehensive transient stability program. Further, rec-
tification equipment is also required in many industrial processes, notably
smelters and chlorine producers, and these are sufficiently large-load items to
warrant good modelling.
The dynamic behaviour of h.v.d.c. links immediately after a large disturbance
either on the d.c. side or close to the converter a.c. terminals requires much more
elaborate models, as discussed in Chapters 11 and 12.
When analysing small perturbations and dynamic stability, it is often assumed
that the converter equipment operates in a controlled manner almost in-
stantaneously when compared with the relatively slow a.c. system dynamics. In
these cases, it is quite acceptable to use a modified steady-state (or quasi-steady-
state) model, the modifications being due to the different constraints imposed by
the load-flow and stability studies. Such a model is also suitable for representing
large rectifier loads during a.c. system disturbances, with further modifications
necessary to represent abnormal rectifier operating modes.
Further to the basic assumptions listed in Chapter 3, the following need to be
made here:
- The implementation of delay angle control is instantaneous.
- The transformer tap position remains unchanged throughout the stability
study unless otherwise specified.
- The direct current is smooth, though its actual value may change during the
study.

Rectifier loads
Large rectifier loads generally consist of a number of series and/or parallel
connected bridges, each bridge being phase shifted relative to the others. With
these configurations, high-pulse numbers can be achieved resulting in minimal
distortion of the supply voltage without filtering. Rectifier loads can therefore be
modelled as a single equivalent bridge with a sinusoidal supply voltage at the
terminals but without representation of passive filters. This model is shown in
Fig. 10.10.
Rectifier loads can utilize a number of control methods. They can use diode
and thyristor elements in full or half-bridge configurations. In some cases, diode
bridges are used with tap changer and saturable reactor control. The effect of the
saturable reactors on diode conduction is identical to delay angle control of a
thyristor over a limited range of delay angles. All these different control methods
 Fig. 10.10. Static rectifier load equivalent circuit

can be modelled using a controlled rectifier with suitable limits imposed on the
delay angle (a).'13'
STATIC LOADS
Operating under constant current control, the d.c. equations are

where A is the constant current controller gain and Ids is the nominal d.c. current
setting as shown in Fig. 10.11.

Natural voltage
characteristic

Constant current
control characteristic

Ids Id
Fig. 10.11. Simple rectifier control characteristic

Constant current cannot be maintained during a large disturbance as a limit of

delay angle will be reached. In this event, the rectifier control specification will
become one of constant delay angle and equation (10.5.2) becomes:
 Protection limits and disturbance severity determines the rectifier operating
characteristics during the disturbance. Shutdown occurs if I, reaches a set
minimum or zero and the voltage Vload will cause shutdown before the a.c.
terminal voltage reaches zero. The action of the rectifier load system is thus
described by equations (3.2.5),(3.2.7),(3.2.lo), (3.2.12), (3.2.13), (10.5.1)and either
(10.5.2) or (10.5.3).

DYNAMlC LOADS

The basic rectifer load model assumes that current on the d.c. side of the bridge
can change instantaneously. For some types of rectifer loads, this may be a valid
assumption, but the d.c. load may well have an overall time constant which is
significant with respect to the fault clearing time. In order to realistically examine
the effects which rectifiers have on the transient-stability of the system, this time
constant must be taken into account. This requires a more complex model to
account for extended overlap angles, when low commutating voltages are
associated with large d.c. currents.
When the delay angle (a) reaches a limiting value, the dynamic response of the
d.c. current (I,) is given by:

where L, represents the equivalent inductance in the load circuit. Substituting for
V, using equation (3.2.5) gives:

where Tdc= L,/R,.

The controller time-constant may also, be large enough to be considered.
However, in transient stability studies where large disturbances are usually being
investigated,faults close to the rectifier load force the delay angle (a) to minimum
very quickly. Provided the rectifier load continues to operate, the delay angle will
remain at its minimum setting throughout the fault period and well into the post-
fault period until the terminal voltage recovers. The controller will, therefore, not
exert any significant control over the d.c. load current. Ignoring the controller
time constant can therefore be justified in most studies.
ABNORMAL MODES OF OPERATION

The slow response of the d.c. current when a large disturbance has been applied to
the a.c. system can cause the rectifier to operate in an abnormal mode.
After a fault application near the rectifier, the near normal value of d.c. current
(I,) needs to be commutated by a reduced a.c. voltage. This causes the
commutation angle (p)to increase and it is possible for it to exceed 60". This mode
of operation is beyond the validity of the equations and to model the dynamic
load effects accurately it is necessary to extend the model.
The full range of rectifier operation can be classified into four modes.(14)
Mode 1-Normal operation. Only two valves in the bridge are involved in
simultaneous commutation at any one time. This mode extends up to a
commutation angle of 60".
Mode 2-Enforced delay. Although a commutation angle greater than 60" is
desired, the forward voltage across the incoming thyristor is negative until
either the previous commutation is complete or until the firing angle exceeds
30". In this mode, p remains at 60" and a ranges up to 30".
Mode 3-Abnormal operation. In this mode, periods of three-phase short
circuit and d.c. short circuit exist when two commutations overlap. During this
period there is a controlled safe short circuit which is cleared when one of the
commutations is complete. During the short-circuit periods, four valves are
conducting. Commutation cannot commence until 30" after the voltage
crossover.

rig. 10.12. Rectifier voltage waveforms

showing different modes of operation.
(a)Mode 1 , ~ ~ 6 0 (b)Mode2,
" ; p=60°
with enforced delay a, ; (c) Mode 3, p > 60"
with short circuit period a,
Mode 4-Continuous three-phase and d.c. short circuit caused by two com-
mutations taking place continuously. In this mode, the commutation angle is
120" and the a.c. and d.c. current paths are independent.
The waveforms for these modes are shown in Fig. 10.12 and Table 10.2
summarizes the conditions
Table 10.2. Rectifier modes of operation

Mode Firing angle Overlap angle

for the different modes of operation. Equations (3.2.5) and (3.2.7)do not apply
for a rectifier operating in Mode 3 and they must be replaced by:

and
a
I , = ---v,e,,(cos a' - cos y')
fixc
Fourier analysis of the waveform leads to the relationship between a.c. and d.c.
current given by equation (3.2.9) where the factor k is now:
3(/ - 2a' - / - 2y' -j2p)
k=
~ ( C Oa'S- COS y')
where

and

A graph showing the value of k for various delay angles (a) and commutation
angles (p) is shown in Fig. 10.13.
Commutation angle (degrees
20 40 60 80 100 120
I I 1 1

1.2-

11-

k 0
-1
.

0.9 -

0.8 -

Fig. 10.13. Variation of k in expression I, = k(3 f l / n ) I ,

IDENTIFICATION OF OPERATING MODE

The mode in which the rectifier is operating can be determined simply by use of a
current factor K I . The current factor is defined as:

Substitution in this, using the relevant equations, yields limits for the modes.
Mode 1:
K I I cos (60" - a)
and
K I _< 2 cos (a) for rectifier operation
Mode 2:

Mode 3:
2.
when a 1 30"

2
K I < -cos(u - 30") when u > 30" (10.5.13)
Jr
Mode 4:
when a I 30"

L
K I = -cos(u - 30") when u > 30" (10.5.14)
Js
This can be demonstrated in the curve of converter operation shown in Fig. 10.14.

Delay angle, a

Fig. 10.14. Converter operation "4)

 It can thus be seen that the mode of operation can be established prior to
solving for the rectifier load equations at every step in the solution.

D.C. link
Provided that it can be safely assumed that a d.c. link is operating in the Quasi
Steady State (QSS) Mode 1, the equations developed for converters in Chapter 3
can be used. That is, the converters are considered to be controllable and fast
acting so that the normal steady-state type of model can be used at each step in
the transient stability study.
The initial steady-state operating conditions of the d.c. link will have been
determined by a load flow and in this, the control type, setting and margin will
have been established.
CONSTANT CURRENT CONTROL

During the solution process at each iteration the control Mode must be
established. This can be done by assuming mode 1 (i.e. with the rectifier on C.C.
control) and by combining equations (3.3.2),(3.5.1)and (3.5.2)a d.c. current can be
determined as:
7

Assuming this current to be valid, then d.c. voltages at each end of the link can be
calculated using equations (3.2.5) and (3.3.2). The d.c. link is operating in mode 2
(i.e. with the inverter on C.C. control) if:
d' ,-1- 'dimodc,<O
The d.c. current for mode 2 operation is given by:

CONSTANT POWER CONTROL

For constant power control, under control mode 1, the d.c. current may be
determined from the quadratic equation:

where Pd, is the setting at the electrical mid-point of the d.c. system, that is:

The correct value for Id-, can then be found from Table 10.3. Control mode 2 is
determined using equation (10.5.16) and in this case the following quadratic
 Table 10.3. Current setting for constant power control from quadratic
equation

1,' 'A, 1,
Within Outside Id I
Outside Within 'A,
Within Within Greater of I,, and I,,
Greater Greater Idms
Greater Less 1,-
Less Greater l~...
Less Less 0

Within =within the range I,,. to

Outside q outside the range I, n, to 1, -.;
Greater = greater than 1, -,
Less = less than 1,

equation is required to be solved.

where

If the link is operating under constant power control but with a current margin
then for control mode 2

It is possible for the d.c. link to be operating in control mode 2 despite

satisfying the inequality of equation (10.5.16). This occurs when the solution
indicates that the rectifier firing angle (a,) is less than the minimum value (a,~,).In
this case the delay angle should be set to its minimum and a solution in mode 2 is
obtained.
It is also possible, that when the link is operating close to the changeover
between modes, convergence problems will occur in which the control mode
changes at each iteration. This can easily be overcome by retaining mode 2
operation whenever detected for the remaining iterations in that particular time
step.

D. C. POWER MODULATION
It has been shown in the previous Section that under the constant power control
mode, the d.c. link is not responsive to a.c. system terminal conditions, i.e. the d.c.
power transfer can be controlled disregarding the actual a.c. voltage angles. Since,
generally, the stability limit of an a.c. line is lower than its thermal limit, the
former can be increased in systems involving d.c. links,'by proper utilization of the
fast converter controllability.
The d.c. power can be modulated in response to a.c. system variables to
increase system damping. Optimum performance can be achieved by controlling
the d.c. system so as to maximize the responses of the a.c. system and d.c. line
simultaneously following the variation of terminal conditions.
The dynamic performance under d.c. power modulation is best modelled in
three separate levels('5).These levels, illustrated in Fig. 10.15, are the a.c. system
controller (i), the d.c. system controller ii and the a.c.-d.c. network (iii).
!)

D.C. system a*Y

D.C. control modes 4 -56 r P!i
L, A.C.1D.C. interface 0, ,0,
(1) ( ii ) ( iii)

Fig. 10.15. A.c.-d.c. dynamic control structure. (i) A.c. system controller ; (ii) D.c.
system controller; (iii) A.c.-d.c. network

A c system
I controller
I

Fig. 10.16. A x . system controller

The a.c. system controller uses a.c. and/or d.c. system information to
derive the current and voltage modulation signals. A block diagram of the
controller and a.c.-d.c. signal conditioner is shown in Fig. 10.16.
The d.c. system controller receives the modulation signals AZ and A E and
the steady-state specifications for power Po current I, and voltage E,.
Figure 10.17 (a) illustrates the power controller model, which develops the
scheduled current setting; it is also shown that the current order
undergoes a gradual increase during restart, after a temporary blocking of
the d.c. link.
The rectifier current controller, Figure 10.17(b), includes signal limits
and rate limits, transducer time constant, band-pass filtering and a voltage
dependent current order limit (VDCOL).
The inverter current controller, Fig. 10.17(c), includes similar com-
ponents plus a communications delay and the system margin current (Im).
 Current order
restart dynamics

IO2
Current rote Current
limit limit Regulator

Current rote Current

lim~t ltrn~t Reaulator

delay

A, A Voltage rate

Fig. 10.17. D.c. system controller. (a) Power controller; (b) Rectifier current
controller; (c) Inverter current controller; (d) D.c. voltage controller

Finally the d.c. voltage controller, including voltage restart dynamics, is

illustrated in Fig. 10.17(d).
(iii) The d.c. current Id and voltage Ed derived in the d.c. system controller
constitute the input signals for the a.c.-d.c. network model which involves
the steady-state solution of the d.c. system (neglecting the d.c. line
dynamics which are included in the d.c. system controller). Here the actual
a.c. and d.c. system quantities are calculated, i.e. control angles, d.c.
current, voltage, active and reactive power. The converter a.c. system
constraints are the open circuit secondary voltages Ear and Eai.

Representation of converters in the network

RECTIFIERS
The static-load rectifier model can be included in the overall solution of the
transient stability program in a similar manner to the basic loads described in
Chapter 9.
 From the initial load flow, nominal bus shunt admittance 6,)can be calculated
for the rectifier. This is included directly into the network admittance matrix [Y].
The injected current into the network in the initial steady state is therefore zero.
In general:

and
y- = -
S*
I Kcrm I
The static-load rectifier model does not depart greatly from an impedance
characteristic and is well behaved for low-terminal voltages, the injected current
tending to zero as the voltage approacheszero. Figure 10.18compares the current

l mpedance

Rectifier (C.C.C.)

Fig. 10.18. Difference between impedance and static

load rectifier characteristic (for VW # 0)

due to a rectifier with that due to a constant impedance load. As the injected
current is never large, the iterative solution for all a.c. conditions is stable.
When the rectifier model is modified to account for the dynamic behaviour of
the d.c. load its characteristic departs widely from that of an impedance.
Immediately after a fault application, the voltage drops to a low value but the
injected current magnitude does not change significantly. Similarly on fault
clearing, the voltage recovers instantaneously to some higher value while the
current remains low.
When the load characteristic differs greatly from that of an impedance, the
sequential solution technique can exhibit convergence problem^,'^) especially
when the voltage is low. With small terminal voltages, the a.c. current magnitude
of the rectifier load is related to the d.c. current but the current phase is greatly
affected by the terminal voltage. Small voltage changes in the complex plane can
result in large variations of the voltage and current phase angles.
To avoid the convergence problems of the sequential solution, an alternative
algorithm has been developed.(13)This combines the rectifier and network
solutions into a unified process. It, however, does not affect the sequential
solution of the other components of the power system with the network.
The basis of this approach is to reduce the a.c. network, excluding the rectifier,
to an equivalent Thevenin source voltage and impedance as viewed from the
primary side of the rectifier transformer terminals. This equivalent of the system,
along with the rectifier, can be described by a set of nonlinear simultaneous
equations which can be solved by a standard Newton-Raphson algorithm. The
solution of the reduced-system yields the fundamental a.c. current at the rectifier
terminals.
To obtain the network equivalent impedance, it is only necessary to inject 1p.u.
current into the network at the rectifier terminals while all other nodal injected
currents are zero. With an injected current vector of this form, a solution of the
network equation (9.5.1) gives the driving point and transfer impedances in the
resulting voltage vector.

where

[I!'nJ.] = and 1; = 1 +j O (10.5.28)

The equivalent circuit shown in Fig. 10.19 can now be applied to find the
rectifier current (I,) by using the Newton-Raphson technique.
The effect of the rectifier on the rest of the system can be determined by
superposition :

where
[m = [Val + [Z] I, (10.5.29)

[ P I = [PI - '[rynj] (10.5.30)

and [iYnj] are the injected currents due to all other generation and loads in the
system.
 1 1 Dynamic
element

Fig. 10.19. Equivalent system for Newton-Raphson

solution

If the network remains constant, vector [Z] is also constant and thus only
needs re-evaluation on the occurrence of a discontinuity.
Thus the advantages of the unified and sequential methods are combined. That
is, good convergence for a difficult element in the system is achieved while the
programming for the rest of the system remains simple and storage requirements
are kept low.
The equivalent system of Fig. 10.19contains seven variables (Vterm,1,,0, $, a, Vd
and I,). With these variables four independent equations can be formed. They are
equation (3.2.5) and:

Equation (10.5.31) is complex and represents two equations. Substituting for Vd

and I, using equations (3.2.11)and (10.5.1)reduces the number of variables to five.
A fifth equation is necessary and with constant current control, that is with the
delay angle (a) within its limits, this can be written as:

Equation (3.2.5) suitably reorganized and equations (10.5.31) to (10.5.33)

represent [F(X)] = 0 of the Newton-Raphson process and:

When the delay angle reaches a specified lower limit (ahn), the control
specification, given by equation (10.5.33), changes to:

and equation (3.2.5) is no longer valid. The d.c. current (I,) is now governed by the
differential equation (10.5.5). If the trapezoidal method is being used, this
equation can be transformed into an algebraic form similar to that described in
Chapter 9. Equation (3.2.5) is replaced by:
The variables ka and kb contain information from the beginning of the
integration step only and are thus constant during the iterative procedure.

J2
kb = ( 1 - 2 kc. ka)Id(t)+ -a.ka. +
cos a(t) 2. ka. V,,,,
Vterm(t) (10.5.38)
71Tdc Rd .
Tdc Rd

where

and t represents the time at the beginning of the integration step and h is the step
length.
Commutation angle p is not explicitly included in the formulation, and since
these equations are for normal operation, the value of k in equation (3.2.11) is
close to unity and may be considered constant at each step without loss of
accuracy. On convergence, p may be calculated and a new k evaluated suitable for
the next step.
In Mode 3 operation, the value of k becomes more significant and for this
reason the number of variables is increased to six to include the commutation
angle p. The equations [ F ( X ) ]= 0 for the Newton-Raphson method in this case
are :

Js
-a.
71
V,er,.cos(O fi
- $ ) . f (p)- -a.
71
Vie,,cos a' 3X = 0
+Lid
71
(10.5.41)

Although k can be calculated explicitly, a linearized form of equation (10.5.8)

obtained for a = 30" can be used to simplify the expression. In the range 60" < p
< 120°, the value of k can be obtained from:
f ( p )= 1.01 - 0 . 0 5 7 3 ~ (10.5.45)
where p is measured in radians.
In Mode 4, the a.c. and d.c. systems are both short circuited at the rectifier and
operate independently. In this case the system equivalent of Fig. 10.19 reduces to
that shown in Fig. 10.20. The network equivalent can be solved directly and the
d.c. current is obtained from the algebraic form of the differential equation
(10.5.5).
 Dynamic

element

Fig. 10.20. Rectifier load equivalent in Mode 4 operation

TS programme Rectifier subroutines

1 - I

I
without rectifier
I/ Extract lheverinvoltaqes
Thevenin impedance
vector for rectifier
I -
from [FI -
I - ' 1
- conditions
Calculate new initial
I Calculate K1 if mode 1 and mode
I ooeration
i 1 .
Y
I

I Calculate mismatches
I I
I YES
I
I NO
I Calculate Jacobianfor \I

I given mode of opemtion

I
I
I
[ Iter =
tter.11
7

II
I
II lter > :1 I(

I
I
I
I
I

b
I

Fig. 10.21. Unified algorithm flow diagram

D.C. LINKS
The problems associated with dynamic rectifier loads do not occur when the d.c.
link is represented by a quasi steady-state model. Each converter behaves in a
manner similar to that of a converter for a static rectifier load. A nominal bus
shunt admittance (jj), is calculated from the initial load flow for both the rectifier
and inverter ends and injected currents are used at each step in the solution to
account for the change from steady state calculated from equation (10.5.24). Note
that the steady-state shunt admittance at the inverter Go,) will have a negative
conductance value as power is being supplied to the network. This is not so for a
synchronous or induction generator as the shunt admittance serves a different
purpose in these cases.

Inclusion of converters in the transient stability program

A flow diagram of the unified algorithm is given in Fig. 10.21.(16)It is important to
note that the hyperplanes of the functions used in the Newtonian iterative
solution process are not linear and good initial estimates are essential at every
step in the procedure. A common problem in converter modelling is that the
solution converges to the unrealistic result of converter reactive power gene-
ration. It is therefore necessary to check against this condition at every iteration.
With integration step lengths of up to 25 mS however, convergence is rapid.

10.6. STATIC VAR COMPENSATION SYSTEMS

The use of static VAR compensation systems (SVS) to maintain an even voltage
profile at load centres remote from generation has become common. An SVS can
have a large VAR rating and therefore to consider it as a fixed shunt element can
produce erroneous results in a transient stability study. Also an SVS may be
installed to improve stability in which case good modelling is essential for both
planning and operation.
The model of the SVS shown in Fig. 10.22 is based on representations
developed by C.I.G.R.E. Working Group 31-01.(") The model is not overly
complex as this would make data difficult to obtain and would be incompatible
with the overall philosophy of a multimachine transient stability program. The
SVS representation can be simplified to any desired degree however by suitable
choice of data.
The basic control circuit consists of two lead-lag and one lag transfer function
connected serially. The differential equations describing the action of the control
circuit with reference to Fig. 10.22 are:

Although electronically produced, the dead band may be considered as a

 Movement

Movement

Fig. 10.23. Dead band analogy and effect. (a) Physical ana-
logy of dead band; (b) The effect of a dead band on output

physical linkage problem as shown in Fig. 10.23(a).In this example, the input (x)
and output (y) move vertically. The diagram shows the initial steady-state
condition in which x and y are equal. The input (x) may move in either direction
by an amount Db/2 before y moves. Beyond this amount of travel, y follows x,
lagging by Db/2 as depicted in Fig. 10.23(b). The effect of a dead band can be
ignored by setting D, to zero.
Stepped output permits the modelling of SVS when discrete capacitor (or
inductor) blocks are switched in or out of the circuit. It is usual to assume that all
blocks are of equal size. During the study, the SVS operates on the step nearest to
the control setting. Iterative chattering can occur if the control-system output
(B,) is on the boundary between two steps. The most simple remedy is to prevent
a step change until B, has moved at least 0.55 B, from the mean setting of the
step.
The initial MVAR loading of the SVS should be included in the busbar loading
schedule data input. However, it is possible for an SVS to contain both
controllable and uncontrollable sections (e.g. variable reactor in parallel with
fixed capacitors or vice versa). It is the total MVAR loading of the SVS which is,
therefore, included in the busbar loading. Only the controllable part should be
specified in the SVS model input and this is removed from the busbar loading
leaving an uncontrollable MVAR load which is converted into a fixed
susceptance associated with the network.
Note that busbar load is assumed positive when flowing out of the network.
The sign is therefore opposite to that for the SVS loading.
 2 0 MVAR

3 t
(a)
7 0 MVAR
( limits 3 0 and 1 0 0 MVAR

30MvARf 1480MVAR
( Limits 0 and 7 0 MVAR lt

Fig. 10.24. Example of an overall SVS controllable and uncontrollable sections.

(a) Example of overall SVS using controllable capacitors; (b) Alternative to overall
SVS in (a) using a controllable reactor

In order to clarify this, consider an overall SVS operating in the steady state as
shown in Fig. 10.2qa). The busbar loading in this case must be specified as
- 50 MVAR and it may be varied between - 10 MVAR and - 80 MVAR
provided the voltage remains constant.
The SVS may be specified in a variety of ways, some more obvious than others,
the response of the system being identical. Three possible specifications are given
in Table 10.4. In the first example, the network static load will be 20 MVAR +
while in the second case the static load will be zero. The third example may be
represented by an overall SVS as shown in Fig. 10.2qb).
Table 10.4. Examples of MVAR loading specification
for SVS shown in Fig. 10.24(a)

Initial Maximum Minimum

Example loading limit limit

It is convenient, when specifying the initial steady-state operation, to use

MVAR. However, this is a function of the voltage and hence all MVAR settings
must be converted to their equivalent per unit susceptance values prior to the
start of the stability study.

Representation of SVS in the overall system

The initial MVAR loading of the SVS is converted into a shunt susceptance (B,)
and added to the total susceptance at the SVS terminal busbar. During the system
study, the deviation from a fixed susceptance device is calculated (B,) and a
current equivalent to this deviation is injected into the network.
A reduction in controlling voltage Vsvwill cause the desired susceptance B, to
increase. That is the capacitance of the SVS will rise and the MVAR output will
increase.
The injected current (Iinj)into the network is given by:

where

Although not necessary for the solution process, the MVA output from the SVS
into the system is given by:
s = pi;vs
and hence

10.7. RELAYS
Relay characteristics may be applied to a transient stability program and the
effect of relay operation automatically included in system studies. This permits
checking of relay settings and gives more realistic information as to system
behaviour after a disturbance, assuming 100 percent reliability of protective
equipment. Reconstruction of the events after fault occurrences may also be
carried out.
Unit protection only responds to faults within a well-defined section of a power
system and as the faults are prespecified, the operation of unit protection schemes
can equally be specified in the switching data input. Thus, only nonunit
protection needs to be modelled and of these overcurrent, undervoltage and
distance schemes are the most common.

Instantaneous overcurrent relays

Instantaneous or fixed time-delay overcurrent relays are readily modelled. The
operating point of the relay should be specified in terms of p.u. primary current
thus avoiding the need to specifically model the current transformer. However,
the location of the current transformer must be specified, e.g. at busbar A on
branch to B, so that thecorrect signals are used by the relay model. The only other
piece of information required is delay-time (t,,,) between the relay operation time
and the circuit-breaker arc extinction time (t,,).
Initially, the circuit-breaker operating time (t,,) is set to some large value as it
must be assumed that the steady-state current is less than the relay setting. At the
end of each time step (e.g. at time t) the current at the current transformer location
is evaluated and if it exceeds the relay setting, the effectivecircuit-breaker time is
set to:

The integration then proceeds until the time-step nearest to tcb when circuit-
breaker opening is simulated by reducing the relevant branch admittance to a
very small value. Alternatively, the integration step length can be adjusted to
open the circuit at time t,,.
During the period between relay operation and tcbthe simulation of relay drop
off may be desired. In this case, if current falls below a prespecified percentage of
relay setting current then tcb is reset to a large value.

Inverse definite minimum time lag overcurrent relays

The inverse time characteristics of induction disc and similar relays may easily be
included in an overcurrent relay. This may be accomplished by defining several
points on the characteristic and interpolating, but curve fitting is better if a simple
function can be found.
For example, an overcurrent relay conforming to British Standard B.S. 142
would appear to be accurately modelled by defining seven points on the curve as
shown in Fig. 10.25(a). However, when plotted on a log-log graph as in
Fig. 10.25(b),the errors are more obvious and can exceed the accuracy limits laid
down in the Standard if care is not taken. However, acceptable accuracy can be
obtained by using the approximation:

top= 3.0/[log (I)] for 1.1 j I 1 20 (10.7.2)

and
to, = c~ for I < 1.1
where to, is the operating time of the relay for a current of I.
Plug bridge setting (Spb)and time multiplier setting (S,,,,), both measured in per
unit, can be incorporated into the relay characteristic and the relay induction disc
travel (D,)at time t due to a current I, may be determined from the previous travel
at time t - h by:

provided that I, 2 1-ISpb.

For currents less than the definite minimum value the relay may be assumed to
reset by spring action. Assuming that resetting from full travel takes 2s then:

when
 Setting multiple
(a)

Six segment
linear approximation

L I I I I
3 10 30 100
Setting multiple
(b)
Fig. 10.25. Inverse definite mini num time lag overcurrent relay (from BS142
(1966)). (a) Line xale; (b) Logarithmic scale
 Initially, travel is set to zero and relay operation is assumed when D equals or
exceeds 1 p.u. If necessary, the relay operating time may be determined by linear
interpolation backwards over the last time interval.
( D , - 1.0)
t =t- .h
OP (4-4-h)
when
D, 2 1.0
and
D l - , < 1.0
and from this the circuit-breaker operating time is given by:
tCb = t o , + tdel
Many static relays have been designed which conform to mechanical
characteristics but they have also permitted different and more suitable
characteristics to be developed. These may be modelled in a similar manner.

Undervoltage relays
Apart from the fact that the relay-operating current is proportional to primary
voltage and not primary current, these relays should be modelled in the same
manner as instantaneous or fixed time-delay overcurrent relays.

Induction machine contactors

The transient analysis of industrial power systems usually require that many
induction machines are modelled. During a disturbance the voltage levels
throughout the system will fluctuate and may result in the machine being
disconnected from the system, albeit temporarily. Other machines may be on
automatic stand-by to maintain essential services.
It is, therefore, necessary to include models of induction machine contactors in
a transient stability program. Undervoltage protection is usually associated with
the contactors and can be modelled in the normal manner.

Directional overcurrent relay

A directional overcurrent relay requires a voltage signal as well as current. The
relay may operate only when the phase difference of the two signals is within
prescribed limits and all other constraints are satisfied.

Distance relays
As in practice, both busbar voltage and branch current signals are required, from
which an apparent impedance Z,/8, of the system at the relaying point can be
calculated. This is then compared with the relay characteristic to determine
operation.
A typical three-zone distance protection relay is shown in Fig. 10.26. Assuming
circular characteristics, then the settings of the relay may be identified by forward
reach Zrf/ measured in impedance (complex)coupled with backward reach R,
+
expressed as a per unit of forward reach. From this information, the centre ( p jq)
and radius (a) of each of the three circles in the impedance plane can be
established :
a = LZ
2 I/('+ Rb) (10.7.7)
p = iZr/(l - R,) cos Brf
q = i Z r/(1 - R,) sin Orf
In the example in Fig. 10.26, R, for zones 1 and 2 is zero.

Fig. 10.26. Three-zone distance relay characteristic

The equation of the boundary of an operating zone is:

+
(Z-cos8- p)2 (Z-sin8- q)2 - a2 = 0 (10.7.8)
and hence operation is defined when:
Z: - 2ZS(p.cos8, + qssin 8,) + (p2+ qZ- a2)< 0 (10.7.9)
Tomato, lens, quadrilateral or other complex characteristics may be con-
structed by combining several simple characteristics of this type.
Each zone has a fixed time-delay associated with it so that the timing for
circuit-breaker action is the same as that described previously.
Incorporating relays in the transient stability program
Nonunit protection equipment usually only trips the local circuit breaker.
Therefore, it is necessary to create dummy busbars so that a faulted branch can be
switched out correctly. This can be done automatically during the data input
stage in the same manner as described in Chapter 9 for faults located on branches.
Thus, a faulted branch may have several dummy busbars associated with it and
care should be taken to ensure all are adequately identified. Protected branches
which are not directly faulted need not be modelled as accurately and the whole
branch may be removed if the circuit breaker at either end is opened.
Relay characteristics should be checked at the end of each time solution and
reconvergence after a discontinuity. It is not necessary to perform the check at
each iteration however. This reduces the computational effort associated with
relays and permits more complex relay characteristics to be modelled at critical
points in the system.
Induction machine switching should not be simulated by creating dummy
branches which can be removed from the network whenever necessary. While this
is a feasible solution, it is extremely wasteful of computational storage and effort.
A more satisfactory method is to identify the state of the machine, i.e. either
switched in or out, by .a simple flag and when switched out to solve for the
machine with zero stator current and likewise remove its injected current from
the network.
The network, however, usually includes a shunt admittance representing the
machine in the initial steady state. This problem may be overcome by injecting
another current to compensate for this admittance whenever a machine is
switched out. Alternatively, a machine liable to switching need not have its
equivalent shunt admittance included in the network at any time during the
study. This simplifies periods when the machine is switched out, but requires a
different injected current to the usual when switched in.
A minor problem occurs when induction machines, which are initially switched
out of service, are included in the input data. An estimate of the full-load active
power of the machine must be specified so that the load characteristics of the
machine can be adequately defined. Also induction motors on stand by for
automatic start-up must be modelled accurately if sensible run-up simulation is
to be achieved.

10.8. UNBALANCED FAULTS

The models developed so far for transient stability analysis have assumed
balanced three-phase operation even during the fault period. Although three-
phase faults are the most onerous, there are occasions when unsymmetrical fault
conditions need to be analysed. It is possible to develop three-phase models of all
power-system equipment but the development effort plus the extra com-
putational costs restrict this type of program to very simple systems. Unbalanced
fault studies are relatively rare and the unbalance only occurs for a short period of
the study thus the need for a three-phase model is limited, and makes full scale
development unattractive.
A more practical approach is the use of symmetrical components. The
negative- and zero-sequence component system models can be added to the
existing single-phase (positive-sequence) model without major disruption and
can be easily removed when not required.

Negative sequence system

Of the two additional symmetrical component systems the negative sequence is
the easier. It is very similar to the positive-sequence system.
The negative-sequence impedances of the components of the transmission
network and static loads are usually the same as for the positive-sequence
impedances and hence no additional storage is required. Phase displacement in
transformer banks is of the opposite sign to that for the positive sequence. While
phase displacement can be ignored during balanced operation, it must be
established if phase quantities are to be calculated during unbalanced operation.
A simple clock notation with each hour representing 30 degree shift is suitable for
this purpose.
The negative-sequence impedance of synchronous machines is different from
the positive-sequence impedance. The flux produced by negative-sequence
armature current rotates in the opposite direction to the rotor, unlike that
produced by positive-sequence current, which is stationary with respect to the
rotor. Rotor currents induced by this flux prevent it from penetrating deeply into
the rotor. The flux path oscillates rapidly between the positive-sequence d and q
axis subtransient flux paths and the negative-sequence reactance X, may
conveniently be defined as:
X, = (X: +X y 2 (10.8.1)
This reactance is the same as the reactance which represents the machine in the
positive-sequence network. The negative-sequence resistance is given by(18):

where Rr is the rotor resistance. While R, and Ra will differ, the overall difference
between the negative-sequence impedance (2,)and positive-sequence impedance
representing the machine is so small as to be neglected in most cases. Further,
rotating machinery do not generate negative-sequence e.m.f.s and hence there is
no negative-sequence Norton injected current.
Thus, ignoring d.c. equipment, the overall negative-sequence network is
identical to the positive-sequence network with all injected currents set to zero.
Negative-sequence currents have a braking effect on the dynamic behaviour of
rotating machinery. For a synchronous machine, where torque and power may
be assumed to be equivalent, the mechanical breaking power (Pb) is(18):
Pb = I;(R, - Ra) (10.8.3)
which may be added directly into the mechanical equation of motion given by
equation (9.2.4).
A similar expression can be found for negative-sequence-breaking torque in an
induction machine where the speed of the rotor and the negative-sequence
currents are taken into account.

Zero sequence system

The zero-sequence system differs greatly from the other two sequence systems.
The zero-sequence impedance of transmission lines is higher and for a
transformer its value and location depends on the phase connection and neutral
arrangements. Fig. 10.27 shows the zero-sequence models for various typical
transformer connections. By replacing the open circuit of transformer types 2,3
and 4 with a very low admittance, the topology of the zero-sequence network can
be made the same as for the other sequence networks.
The zero-sequence impedance of rotating machinery must be specified in the
data input so that its inverse can be included in the zero-sequence system
admittance matrix. As with the negative-sequence system model, there is no zero-
sequence e.m.f.,s generated and hence there is no Norton injected currents into
this system.

Transformer type Vbrious winding configurations

Type 4
P 0
o--=.o---O

Fig. 10.27. Modelling of zero-sequence equivalent networks of

transformers
Inclusion of negative and zero sequence systems for unsymmetrical faults
The major effect of unsymmetrical faults is to increase the apparent fault
impedance. On fault application, the negative- and zero-sequence impedances of
the system at the point of fault are calculated. These are simply the inverse of the
self-admittances at the point of fault and are determined in the same manner as
described by equations (10.5.27) and (10.5.28). Depending on the type of fault, the
fault impedance is modified to include the negative- and zero-sequence imped-
ance. The fault impedance then remains constant until changed by either branch
switching or fault removal.
If negative-sequence-braking effects are to be included, it is necessary to
evaluate the negative-sequence current in the relevant machines at each iteration.
This is done by injecting the negative-sequence current, determined at the point
of fault, into the negative sequence system admittance matrix [y2].

where [if2]is a zero vector except at the point of fault. The vector [t2]contains
the negative-sequence voltages at all busbars from which the machine negative-
sequence currents are readily obtained.
If phase information is required, then the zero-sequence voltages at all busbars
also need to be determined, depending on the type of fault. This is done in an
identical manner to that used for the negative-sequence system.

10.9. GENERAL CONCLUSIONS

This chapter has extended the capabilities of the transient stability analysis
program developed in Chapter 9. This has been achieved by producing more
advanced models of some basic power-system components and also by
introducing models of less frequently simulated equipment.
A transient stability program need not necessarily contain all the models
described in order to completely describe a power system. Conversely, a program
containing all these refinements is not necessarily adequate for a particular
system. It must be anticipated that transient stability programs will be con-
tinuously refined as tighter operating constraints coupled with new control
strategies are introduced.
Moreover, the h.v.d.c. converter model presented in this chapter has assumed
continuous controllability under all circumstances. A much more accurate model
of converters is developed in the next chapter and the integration of this model
with a multi-machine transient stability program is discussed in Chapter 13.

10.10. REFERENCES
1. S. B. Crary, 1945. Power System Stability-Steady-State Stability, Vol. 1. Wiley &
Sons Ltd., New York.
2. D. W. Olive, 1966. 'New techniques for the calculation of dynamic response', IEEE
Transactions on Power Apparatus and Systems, PAS-85(7), 767-777.
 3. T. J. Hammons, and D. J. Winning, 1971. 'Comparisons of synchronous-machine
models in the study of the transient behaviour of electrical power systems',
Proceedings I EE, 118 (lo), 1442-1458.
4. S. Beckwith, 1937. 'Approximating Potier Reactance', AIEE Transactions, 813.
5. A. H. Knable, 1956. Electrical Power Systems Engineering-Problems and Solutions.
McGraw-Hill, New York.
6. H. W. Dommell, and N. Sato, 1972. 'Fast transient stability solutions', IEEE
Transactions on Power Apparatus and Systems, PAS-91 (8), 1643-1650.
7. C. P. Arnold, 1976. 'Solutions of the multi-machine power-system stability problem.'
Ph.D. thesis, Victoria University of Manchester, U.K.
8. IEEE Committee Report, 1973. 'Dynamic models for steam and hydro turbines in
power-system studies', IEEE Transactions on Power Apparatus and Systems, PAS-92 .
(6), 1904-191 5.
9. D. S. Brereton, D. G. Lewis, and C. C. Young, 1957. 'Representation of induction
motor loads during power-system stability studies', AIEE Transactions on Power
Apparatus and Systems, PAS-76, 45 1-461.
10. H. E. Jordan, 1979. 'Synthesis of double-cage induction motor design', AIEE
Transactions on Power Apparatus and Systems, PAS-78, 691-695.
11. C. P. Arnold, and E. J. P. Pacheco, 1979. 'Modelling induction motor start-up in a
multi-machine transient stability program.' IEEE PES Summer Meeting. Vancouver,
B.C., Canada.
12. E. J. P. Pacheco, 1975. 'Induction motor starting in an electrical power-system
transient-stability programme.' M.Sc. dissertation, Victoria University of
Manchester, U.K.
13. C. P. Arnold, K. S. Turner, and J. Arrillaga, 1980. 'Modelling rectifier loads for a
multi-machine transient-stability programme', IEEE Transactions on Power
Apparatus and Systems, PAS-99 (I), 78-85.
14. D. B. Giesner, and H. Arrillaga, 1970. 'Operating modes of the three-phase bridge
converter', Int. J. Elect. Engng. Educ., 8, 373-388.
15. IEEE Working Group on Dynamic Performance and Modeling of DC Systems, 1980.
'Hierarchical Structure'.
16. K. S. Turner, 1980. 'Transient stability analysis of integrated a.c. and d.c. power
systems', Ph.D. Thesis. University of Canterbury, New Zealand.
17. CIGRE Working Group 31-01, 1977. 'Modelling of static shunt VAR systems for
system analysis', Electra, (51), 45-74.
18. E. W. Kimbark, 1956. Power System Stability: Synchronous Machines. Wiley & Sons,
New York.
 Dynamic Simulation of
A.C.-D.C. Systems
11.1. INTRODUCTION
During disturbances, the interaction between supply systems and large power
converters, especially those used in h.v.d.c. transmission, generally results in large
waveform distortion and valve maloperations. Such behaviour cannot be
predicted with the quasi-steady-state models described in Chapter 3, because the
topological converter changes resulting from the disturbance can not be specified
in advance.
The simulation of transients associated with h.v.d.c. transmission is presently
carried out in physical models and simulators which provide the required
detailed modelling of the converter plant.
However, the pressing need to represent the behaviour of h.v.d.c. converters in
conventional a.c. power-system computer studies is exciting the development of
digital methods. Such methods are normally limited in their practical application
by the restricted size and accuracy of the a.c. system representation and by the
cost of running the programs. As a result, programs designed to analyse the
transient behaviour of static converters have tended to use a simplified a.c.
network representation.
This chapter describes a general dynamic model, formulated in terms of state-
space theory. It uses nodal analysis with diakoptical segregation of the plant
components undergoing frequent switching, to avoid involving the whole
network in unnecessary topological changes. The model accepts varying degrees
of a.c. system and converter plant representation to meet the requirements of any
particular study, e.g. small disturbances, faults, etc. In each case the individual
units of greater relevance are emphasised, whereas the rest of the power system is
synthesized into a simpler equivalent circuit.

11.2. FORMULATION OF THE DIFFERENTIAL EQUATIONS

As explained in Chapter 2, a power system can be represented by a network of
inductive, resistive and capacitive elements. The behaviour of the network, i.e. the
currents in the different branches and the voltages at different nodes, is
determined by topological and algebraic constraints.
In cases involving h.v.d.c. links and containing few capacitive branches, mesh
analysis results in fewer current differential equations. However, when the a.c.
system contains multi-conductor transmission lines with mutual inductive effects
and a high level of interconnection, the nodal approach is more efficient.
Moreover, the accessibility of individual elements is important in a flexible
model, e.g. to include nonlinearities such as saturation, and this is achieved by
using the branch current formulation.

Definitions
Let the network contain n nodes interconnecting 1inductive branches, r resistive
branches and c capacitive branches subject to the following restrictions('):
- one end of each capacitive branch (or subnetwork) is the common reference
point of the system;
- at least one end of.each resistive branch (or subnetwork) is the common
reference or a node also having a capacitive connection;
- at least one end of each inductive branch (or subnetwork) is the common
reference or a node having a resistive or a capacitive connection.
These restrictions are taken care of in the representation of the system
components.
The network nodes can thus be subdivided into three parts according to the
types of branches connected to them namely;
- a nodes: which have at least one capacitive connection.
- nodes: which have a least one resistive connection but no capacitive
connections.
- y nodes: which have only inductive connections.

The topological matrices Kf,, K:,, K:, which are the branch-nodes incidence
matrices (transposed) of the 1, r and c branches, respectively, are defined by their
general elements as follows:
K;, = 1, if node i is the sending end of branch p
= - 1, if node i is the receiving end of branch p
= 0, otherwise.
These matrices can be partitioned according to node type as follows:
Kfn = CKfaKfsKf,l
Kin = CK:aK:BK: ,I (11.2.1)
Kin = CK:aKBK:
where K:,, K:s and K:, are null by definition.
302
Voltage and current relations
In the absence of current sources, the following nodal equation applies

KnlIl+ K n J , + KWIC= 0 (11.2.2)

where I are the branch current vectors.
Partitioning equation (11.2.2) according to equation (11.2.1) and rearranging:
KyIIl= 0
or
K Y ~ P I I 0= (11.2.3)
+ Kp1Il= 0
KJ, (11.2.4)
KmIc+ Karlr+ KalIl= 0 (11.2.5)
The following branch equations can be written as matrix expressions for each
branch type.
(a) Inductive branches
+ +
El - p(LIIl)- RIIl K &Va KKf V p KKf, Vy= 0 +
or
+ +
pIl = L r 1 ( E 1-pLl.Il - RIIl K&Va KIBVp K[,VY) +
(b) Resistive branches
- R,I,+ KLVa + K:,Vp =O
or
I, = R,- ' ( K ; va+ K ; vp)
~
(c) Capacitive branches
Ccp(K;aVal = Ic
Premultiplying by K , and substituting equation (11.2.5) yields
KacCcK:a~Va= - Ka1I1- Karl,
or
pVa=CilJa
where
C,' = K,CcK:a

and
Ja = - KalIl - Karlr
= K,Ic
Evaluation of dependent variables
Premultiplying equation (11.2.9) by Kg, yields
+
V, = - RB(K,,Il K,,R; 'KL VJ (11.2.15)
where
R, =(K,,R,-'K~,)-~ (11.2.16)
Premultiplying equation (11.2.7) by K,, and substituting equation (1 1.2.3)
yields

where

Equations (11.2.7)and (11.2.12)can now be integrated by numerical techniques

to assess the system behaviour. The vectors V, (equation (11.2.15)),V, (equation
(11.2.17)) and I, (equation (11.2.9))may be eliminated entirely from the solution
so that only I,, Vu and the input vectors are explicit in the equations to be
integrated. However, although this elimination reduces the overall number of
equations, it has the following computational disadvantages.
- some of the original matrix sparsity is sacrificed;
- the modified incidence matrices now contain noninteger elements, thereby
requiring actual multiplications rather than simple additions and sub-
tractions when performing connection matrix-by-vector products;
- in addition to modifying L, and K,, at every converter topological alteration,
an auxiliary branch matrix is created in the elimination process, which also
need reforming at these times.
It is, therefore, recommended to retain the equations in the above form and to
evaluate the dependent variables V,, Vy and I, at each step. Moreover, this
formulation provides useful output information directly.

Formulation for variable topology

In Fig. 11.1, the converter transformer windings and the d.c. smoothing reactor
together with the bridge valves, which are considered as switches, form a
subnetwork of inductive elements whose topology varies periodically with the
changes in valve states. Such a network contains only y-nodes and the variable
topology is manifested by changes in L, and K,,.
To avoid involving the whole network in a topological change affecting on14
part of it, it is generally convenient to use an extra type of branch (k) and node (6)
to represent the part of the system subject to the frequent topological change, i.e.
the converters.
+
With that addition, the inductive subnetwork contains ( I k) inductive
 Convertor transformer
and o.c, system

I dl

1" "3
0 >= *>b -
and d.c. system
Ys

d2
Fig. 1 1 . 1 . Injected currents

branches; the k branches denoting branches involved in the converter repre-

sentation, and the 1 branches denoting all inductive branches external to and not
connected directly to the converter. An inductive branch between the two
inductive subnetworks will thus be specified as a k branch so that the connection
matrix Kf, is null. Therefore, 6 nodes connect only inductive branches within a
converter subnetwork (k branches) whereas y nodes also involve external
inductive (I) branches. Therefore:
Kin = CK:,Ki&K$I (11.2.19)
and K',,, K:,, Kz, are all null.
Because of the nature of the converter subnetwork, it may be assumed that it
contains no e.m.f. sources, or time variant inductances (i.e. pLk = 0). Applying
partitioning to Kirchhoffs current law, yields:

The k branch voltage equations may be written as

+
pIk = L , '( - RkIk Kf, Va + KipV, + KiYVy+ K$ V,) (11.2.24)
along with three equations from the previous section
 305
and from equation (11.2.23) the latter can also be written as
pVa = c- iK a l z i - K a k z k - K a r z r ) (11.2.28)
The three dependent variables VD,V yand V, can now be evaluated directly as
follows :
Prernultiplying equation (11.2.9) by KBr,yields
' +
VD= - RD(KDrRr-K:, Va KDIIl KDkIk) + (11.2.29)
Using equation (11.2.20) and multiplying equation (11.2.24) by K,,, yields
0 = K,,L; '( - R,I, + K I , va+ K;D VD+ K;, V, + K;, V*)
Then defining
L, = (K,,L, ' K i d ) - '
yields
' + +
VJ= - L,KJkL; ( - RkIk K', Va KkDVD+ K;, V,)
Differentiation of equation (11.2.21) yields
Kdk~zk + K J l ~ I=l 0
Substituting equations (11.2.24) and (11.2.7) yields
K,& +
'( - RkIk K;, va+ Kfka VD+ K;, V, + K;, V*)
+ K,,I; ' ( E l- RIIl- pL,Il + K$ Va+ KfDVD+ K:, V,,)= 0
or
'( - RkIk+ K I , va+ K& VD+ Kid V*)
vy= - LLa[KykLk
+ K,,I; '(El - RIIl- pLl.Il + Kia Va+ KiDVD)]
where

which is not affected by topology changes, but is, in general, not sparse.
Inspection of equations (11.2.32) and (11.2.35) indicates that they are two
interdependent equations for the two vectors Vyand V,. To solve for these two
vectors would require an iterative process, but the associated computational
burden would be prohibitive. However, in a practical interconnected a.c.-d.c.
power system, the only interface between converter and nonconverter inductive
subnetworks occurs at the a.c. terminals of the convertor transformers. The usual
practice is to have a.c. harmonic filters, including a high pass, connected on the
a.c. side, which would by definition create a /?node at this busbar. Alternatively,
the connection of an a.c. line or static capacitors will define it as an a node. Thus, if
the restriction K,, = 0 is applied, the computational problem is removed without
imposing any real system representation restrictions. In the rare case of the
converter busbar being a y node, the formulation without 6 nodes would be more
economic.
By applying this network restriction, i.e. K y k null, to the above equations, a
direct solution is obtained, i.e.
v, = - L,KdkL; '( - RkIk+ Kia Va+ K:. v8) ( 1 1.2.37)
and
Vy= - L y K y l L ;'(El- RIIl- pLl.I1+ K h Va+ KfS Vg) (11.2.38)
Matrix L, in equation ( 1 1.2.37)is block-diagonal with the blocks correspond-
ing to node clusters which are separated by non-&nodes. This implies that a
switching operation in a converter will cause a change in only that part of L,
where the switching takes place.
As already explained, the primary windings of the converter transformers are
normally connected to either a or P nodes, and therefore these transformers can
be considered short circuited at their primary terminals for the purpose of setting
up L, or modifying it. For this purpose, the transformer phases can be simply
represented by their leakage reactances referred to the secondaries and connected
to the corresponding secondary winding ends.
The modification of L, to represent the various modes of operation can be
effected using a simple procedure which adds or removes single elements from L,
thereby avoiding re-formation and inversion for each mode.
When an additional link of inductance ( L )is connected between nodes i and j in
the network, L, is modified to the new matrix LL whose general element is given,
in terms of the elements of L,, by

where
+
S = Lii L j j - L i j - L j i + L
One of the nodes i and j may be external to the &network, in which case the
corresponding elements in equation ( 1 1.2.39)will be zero. Conducting valves are
represented by adding a link of zero inductance and the removal of an existing
branch is represented by adding a link of negative inductance equal in magnitude
to that of the removed branch and connected in parallel with it. In the latter case,
if the branch to be removed provides the only path from one part of the network
to the external system or to the common reference point, that part of the network
will become isolated. This condition is indicated by a vanishing denominator (S),
and it must be avoided.

11.3. TRANSIENT CONVERTER MODEL

The converter bridge contains the three terminals of the transformer secondary
windings and the two d.c. terminals. One of the d.c. terminals may be earthed, or
may be connected to a d.c. terminal of a second bridge. A reduction in the number
of nodes together with a considerable improvement in the efficiency of the
treatment can be achieved if the following assumptions are made:
(i) The forward voltage drop in a conducting valve is neglected so that the
valve may be considered as a switch. This is justified by the fact that the
voltage drop is very small in comparison with the normal operating
voltage, quite independent of the current and should, therefore, play an
insignificant part in the commutation process, since all the valves
commutating on the same side of the bridge suffer similar drops.
(ii) The converter transformer leakage reactances as viewed from the
secondary terminals are identical for the three phases. This provides a
useful simplification since it enables the use of the same nodal solution
matrix (La) for any combination of conducting valves within the same
mode. This assumption is not justified if nonlinearities due to saturation
are considered but it is perfectly legitimate otherwise.
(iii) The case of a completely isolated transformer secondary is not to be
allowed, i.e. at least one of the bridge valves is kept O N when the converter
is in a nonconducting state. This assumption is necessary for meeting the
restriction on inductive branches imposed by a state variable formulation,
i.e. the voltage of one of the terminals must be fixed using a conducting
valve. Clearly, this will not influence the remainder of the system since the
transformer secondary remains open.

Modes of operation
With reference to the three-phase bridge configuration of Fig. 11.1, the
interconnection of the converter transformer secondary windings and the d.c.
smoothing reactor, is determined by the number and positions of conducting
valves. For the converter bridge to be in a conducting state, two or more valves
must be ON in such a way that they provide a closed path for the current as
viewed from the d.c. terminals. With six valves in the bridge, conducting states
can be categorized into the following modes:
A- (Noncommutating mode): two valves on different sides and arms of the
bridge are ON.
B- (Normal commutation): three valves, one on each arm, two commutat-
ing and at least one conducting on each side of the bridge.
C- (Noncommutating arm short circuit): both valves on one of the arms are
ON, with no other valves conducting.
D- (Commutating arm short circuit): as in C with one or both valves on
another arm being ON. With one valve on the second arm conducting
(Dl), there is a single commutation process. When both valves on the
second arm conduct (D2), commutations are simultaneously taking
place on both sides of the bridge.
E- (A. C. short circuit): this is a case when the three secondary terminals of
the converter transformer are short circuited by the conducting valves.
 This occurs when four or more conducting valves involve all three arms
of the bridge. With four conducting valves (El), two commutations occur
simultaneously which may be on the same side of the bridge. When five
valves conduct (E2),commutations take place on both sides and with six
valves conducting (E3), multiple commutations take place on both sides
of the bridge.
The bypass valve, used in many existing schemes, has been left out of the above
classification since its effect is implicitly included in Modes C, D and E.

Converter topology
The modes of operation stated above are represented by defining a set of three
nodes for each converter bridge nominally referred to as the top (a,), middle (6,)
and bottom (6,) as shown in Fig. 11.2. 6, and 6, correspond to dl and d,,
respectively, in Fig. 11.1 during normal operation and the presence of conducting
valves determines the connections of the transformer secondary terminals (a, b
and c) to any of the three nodes 6,, 6, and 6,.

Fig. 1 1.2. Converter bridge 6 nodes

Referring to Fig. 11.2, if valve v, is ON then a is set to 8, and if valve v4 is ON

then a is set to 6,, otherwise, if both v, and v4 are OFF, a is set to 6,. When the
process is repeated for b and c, all bridge nodes (a, b, c, dl and d,) will have been
assigned to the set of 6-nodes defined above.
Since the connections of the transformer secondary windings and the d.c.
smoothing reactor are permanently fixed to a, b, c, d, or d,, then the connection
matrix K,, is fully established by the procedure defined above. There are,
however, certain instances when two phases simultaneously have no conducting
valves and will, therefore, both be set to 6,, such as when operating in Mode C. In
these cases, the smoothing reactor bridge terminal (dl or d,) is set to the opposite
side of the bridge, describing a short circuit, or to the other end of the reactor,
describing an open circuit, as the case may be. The latter case is permissible since
the smoothing reactor normally has no mutual coupling with any other inductive
element in the network. Moreover, if the third phase is also assigned to the node
which normally corresponds to the smoothing reactor bridge end, a similar
change can be carried out for that phase. This process leaves a free node to which
one of the two open phases may be assigned, thereby establishing the
independence of these two phases.
The nodal solution matrix L, is established for the mode of operation in
question, taking into consideration the assumptions and characteristics dis-
cussed in the preceding sections, as follows:
Defining a 'basic' L, as one corresponding to the noncommutating mode
(Mode A), a 'working' L, is derived in accordance with the actual mode of
operation by adding, removing or changing the connections of the appropriate
inductive elements using equation (11.2.32). For example, when the bridge
operates in the normal commutating mode (Mode B), with the commutation
process taking place in the top half of the bridge, e.g. valves u,, u, and u, conduct,
a zero inductance link, representing an extra valve conducting in comparison
with the 'basic' mode, is inserted between 6, and 6, according to equation
(11.2.32).
The above treatment is quite general and applies equally well to all the
configurations shown in Fig. 11.3. Details of node numbering, formation of the
'basic' L, and its modification for the modes of operation are considered in the
next two sections.

6- Nodes and basic L ,

All the converter configurations shown in Fig. 11.3 assume that one end of the
bridge combination is earthed and the other connected to the smoothing reactor.
Hence, only two nodes are explicitly defined in six-pulse operation to simulate

Nominal
inverters
-
(d (e

Fig. 1 1.3. Different converter configurations

Table 1 1 . 1
- -- --

Nominal rectifier &range

Nominal inverter
Type of converter Neutral 6, 6, 6, (topearthed) min max

6-pulse STAR m+l m+2 m+3 0 61-6, m+l m+3

6-pulse DELTA - m+l m+2 0 6, -8, m+l m+2
m+l m+2 m+3 m+4 4-6, m+l m+5
- m+4 m+5 0
4-6, m+l m+5
valve switching, with a further node needed to represent the neutral of the
transformer secondary in a star bridge. Five nodes altogether are needed to
simulate twelve-pulse operation of which one is shared between the two bridges
and another representing the neutral of the star-connected secondary. The
allocation of new &nodes to a new set of converter data, assuming that m such
nodes have already been defined, is shown in Table 11.1.
Table 11.1 indicates that a fixed structure is used in representing the variation
in topology caused by valve switchings. This implies that, during commutations
for instance, redundant nodes are retained to the detriment of computation
efficiency. However, the dynamic reordering and relocation of data, which would
otherwise be required, is avoided by adopting the fixed structure. The converter
bridge, on the other hand, should, under normal conditions, be in a noncom-
mutating state, which requires all the nodes defined, for the greater part of the
time. In some special cases, explicitly the noncommutating arm short-circuit and
the open-circuit condition, the number of nodes defined above is insdlicient to deal
with these cases in the direct manner used in the others. Action is taken in these
special cases to free one of the redundant nodes and use it to represent the
condition as described in the next section. Thls approachis preferred, to deal with

Fig. 1 1.4. Noncommutating nodal matrices. (a) Star ;

(b) Delta; (c) Twelve-pulse
the special conditions since the addition of extra nodes would considerably
increase the computation time.
The basic noncommutating solution matrix (La) is established a t input by a
direct procedure in which each secondary winding of the converter transformer is
replaced by its equivalent leakage reactance and the smoothing reactor is
excluded. The procedure results in the matrix shown in Fig. 11.4. An inductance
+
x,, representing the smoothing reactor, is then connected between m 1 and 0 for
a six-pulse delta converter or between m + 2 and 0 for six-pulse star and twelve-
pulse converter using equation (11.2.32). The equivalent leakage reactance
referred to above is derived from the inductance matrix of the transformer and is
given by :
x, = L22 - L21 .Ll2ILll
The matrices given in Fig. 11.4 can be verified by inspection.

Representation of modes of operation

In normal operation, converter bridges alternate between the noncommutating
and commutating modes, i.e. Modes A and B defined earlier. The nodal solution
matrices used in these cases are, respectively, the unmodified basic La and the one
modified by connecting a zero inductance link between 6, and either 6, or a,,
depending on where the commutation process is taking place. The commutating
short-circuit condition (Mode D l ) is also easily represented by connecting a
single zero inductance link between 6, and 6,. The three-phase short-circuit
(Mode El) is represented by connecting a zero inductance between 6, and a,, and
another between 6, and 6,.
However, in the noncommutating d.c. short circuit (Mode C), due to the
limited number of 6-nodes available, special action is taken to represent the
conditions. With six-pulse operation the smoothing reactor is removed from its
normal 6-node and connected to the common reference point. This action frees a
node to which one of the two nonconducting arms may be assigned.
Twelve-pulse converters can be more complicated, depending on which of the
two bridges is in Mode C. If the short circuit is in the star bridge, the reactor is
removed from its normal 6-node and connected to the node shared between the
+
two bridges (m 4 in Table 11.1), as shown in Fig. 11.5(a). If, however, the delta
bridge is in Mode C, then all the branches in the star bridge that are connected to
+
(m 4) are disconnected from this node and connected to the common reference
point. The branches switched over are effectively connected in parallel and the
program lumps them together and, therefore, one removal and connection
process is required. Fig. 11.5(b) shows this condition with the star bridge in a
commutating state on the shared-node side. Finally, if both bridges are in
Mode C, the smoothing reactor and one transformer branch in the star bridge are
both disconnected from their normal 6-nodes and connected to the common
reference point as shown in Fig. 1 1 3 ~ ) .
To represent the converter when it is in a nonconducting mode, i.e. the current
 Fig. 1 1.5. Noncommuting arm short circuit in twelve-pulse
converters

path is interrupted, the modification of La required is to remove the smoothing

reactor from its normal b-node and connect it to an extra b-node (6,) whose
voltage is set equal to the d.c. line voltage. This frees a node which can be assigned
to any of the transformer secondary terminals. The extra node 6, is not
represented in La and, therefore, the structure of the converter network is
unchanged by its definition. The condition arrived at describes a true open circuit
on the smoothing reactor side of the converter.
In all the cases described above, the matrix La is modified by the switching
actions indicated according to equation (1 1.2.31). Moreover, since the modifi-
cation will only change the elements of La corresponding to the nodes of the
wnverter in which the switching has taken place, the modification is limited to
the parts of La ranging between the minimum and maximum node numbers
defined in Table 11.1.
The K,, incidence matrix is formed for the specific combination of conducting
valves as described briefly in the section on converter topology. Fig. 11.6 shows, ,
as an example, a twelve-pulse converter in which the A-bridge is undergoing
commutation. The converter b-nodes are assigned to the bridge and transformer
terminal nodes, according to the combination of conducting valves given, as
 Bridgenodes n d
o, b , cl dl 2 o2 b2 c2 d3 d4
(b)
%-nodes 8, S2 B3 B4 82 B4 B4 B4 0 0
a4

Fig. 1 1.6. Formation 6 - k incidence matrix

shown in Fig. 11.6(b). The permanent branch list shown in Fig. 1 1 . q ~ )is
transformed to the working list, defined in terms of &nodes shown in Fig. 11.6(d).
Finally, the K,, incidence matrix, shown in Fig. 11.6(e), is derived from the
working branch list. The process is limited to the bridge nodes, and the external
connections of the converter network, established permanently at input, are
indicated by zero entries.

Valve voltages and currents

Referring to Fig. 11.1, the forward voltage across the top and bottom valves of
arm p are (Vp- Vd,) and (vd2- Vp) respectively. Since terminals p, d l and d2
correspond directly to the 6-nodes defined in the section on converter topology,
whose voltages can be evaluated using equation (11.2.29),the calculation of valve
 Fig. 1 1.7. Injected currents

voltages becomes a matter of routine. The relations stated will apply regardless of
whether the valve is conducting or not, for, when the valve is conducting, its end
will be set to the same b-nodes and its voltage will correctly be calculated at zero.
The current relations, on the other hand, are entirely determined by two
factors: the currents in the converter branches, i.e. the smoothing reactor and the
transformer windings, and by the prevailing combination of conducting valves.
Considering the set of injected currents shown in Fig. 11.7, conducting valves
provide the paths for these currents to flow in the system and the instantaneous
values of valve currents can be obtained from these injected currents. The injected
currents themselves are obtained directly from the branch currents. J,, and J,,
are equal and numerically equivalent to the d.c. current. The phase injected
currents are obtained from the current in the transformer secondary windings,
e.g. for a star-connected secondary, J, is the negative of the current in the
secondary winding connected to node a and for a delta-connected secondary, the
injected currents are given by the difference between the currents in the windings
connected to the node.
The relationship between the valve currents in the top and bottom sides of the
bridge (I,, and I,,) on arm p, with the injected current at p, Jp, is
Jp =I v l -I v b
when the bottom valve (vb) is OFF:

and when the top valve (vt) is OFF:

In these two cases, an implied supplementary relation, setting the nonconduct-

ing valve current to zero, enables the solutions given by equations (11.3.2) and
(11.3.3) to be obtained. However, when both valves conduct the supplementary
relation must be found elsewhere. The method suggested considers the three
valves on the top half of the bridge, whose currents add up to J,,, i.e.:
in this relation, provided p is the only arm with both valves conducting, I,, is
implicitly defined.
Equation (1 1.3.4) may be rewritten, substituting by equation (11.3.2) for the top
conducting valves on phases other than p, to give

where q are the phases having their top valves conducting.

The current in the bottom valve of phase p may now be calculated using
equation (11.3.1), or by a relation similar to equation (11.3.5) giving

where r are the arms having their bottom valves conducting.

It can be seen that no solution is possible when two bridge arms have both
valves conducting since there is an insufficient number of independent equations.
A further relation would be provided by considering the current which may
circulate in the closed loop formed by the two arms but this entails assumptions
which may not be justifiable. In practice, the case is likely to arise only in the
special circumstance when two valves are fired simultaneously at a time when
another pair of valves, which would close the loop, are conducting. Otherwise,
there is insufficient voltage across a further incoming valve, if the three other
valves are already conducting and a firing pulse is applied to the fourth, for it to
conduct. For this reason the operating Modes D2, E2 and E3, defined in the
earlier section have been left out.

Initial conditions
Conducting valves in each bridge are determined from knowledge of phase 'a'
voltage phase angle 8,, measured with respect to the steady-state load-flow
reference, and taking into consideration the inherent phase shift in the bridge
+
transformer 8, (0 or 30") and the delay angle a.
RepreSenting the a.c. voltage as a cosine function, the voltage crossover for
valve vl in the bridge, at time t = 0, will have occurred 'A' degrees earlier.
Referring to Fig. 11.8, A is given by:

Valve vl will, therefore, have started conducting B degrees earlier, where

In this condition valves v, and v, are conducting and valve v2 is due to be fired at
time F,,, which is 60 degrees later than the firing of v,, hence,

However, if B is less than zero then v, is yet to be fired. Conversely, If B is

 Conducting Winding currents (Id = 3)
valves a I b 1 c

Fig. 11.8. Initial conditions in converter bridges. (a)

Determination of conducting valves; (b) Currents in
star bridge; (c) Currents in delta bridge; (d) Delta
winding currents

greater than 60 degrees then v , will have been conducting for more than one firing
period indicating that u, is already conducting. In either case, the sequence is
changed until the correct pair of conducting valves has been defined. This
approach ignores the commutation angle which, if included, would probably
complicate the issue without offering a substantial advantage since the initial
system conditions are approximate in the first place.
A refinement is to adjust the load-flow reference angle so that commutation
periods are avoided at the converter. This can be effected by choosing the firing
time of the next valve, k + 1, to be
 317
where in the above analysis k = 1. Then if the steady-state commutation angle
is p.
C = 60 - p for six-pulse operation
and
C = 60 - 8, - p for twelve-pulse operation
This will ensure maximum period exists before any alteration in the converter
topology is effected by a valve commutation. Therefore
B=k.60-C
and
A=k.60+a-C
and the required value for 8, will be
&P=(k- 1 ) . 6 0 + 8 , + a - C
which can be achieved by altering the load-flow reference angle by ( e P- 8,)
Having established the pair of conducting valves in the bridge, the currents in
the transformer windings must be established on both sides. For a star bridge,
shown in Fig. 11.8(b), the secondary winding currents of each phase are set equal
to the given d.c. current if its bottom valve is conducting and to the negative of
that if the top valve is conducting, otherwise it is set to zero. Primary (a.c. side)
winding currents are set equal to the corresponding secondary currents with the
signs reversed.
Current is forced to flow in a secondary winding of a delta bridge even though
both valves corresponding to its phase are not conducting. In the case shown in
Fig. 11.8(c),the secondary winding of phase (c) carries a third of the d.c. current
with valves v, and v , both not conducting. In general, one winding carries two-
thirds of the d.c. current in one direction while the other two carry one-third each,
flowing in the opposite direction. The table shown in Fig. 11.8(d)gives the current
distribution in the delta winding for all combinations of conducting pairs of
valves. The primary winding currents are set to ( - fi) times the corresponding
secondary currents.

Converter control system

In order to illustrate the representation of control schemes in the context of full
dynamic simulation a model based on the equidistant firing") control system is
described here.
A valve is switched to the conducting state when there is adequate positive
voltage between anode and cathode and a firing pulse is present. The next valve to
fire is then set by a ring counter in the sequence indicated in Fig. 11.1 and its firing
instant is given by:
where APi is the corrective action produced by the control system which in the
steady state is zero.
With rectifier operation under constant current control (C.C.)

where K, and K, are adjustable constants.

In practice, the C.C. action is somewhat more complex, e.g. the constant
current amplifier in the phase-locked oscillator system is provided with a
feedback damping circuit. In essence the factors K, and K,, corresponding to the
effective gain of the system, would be complex time functions representing the
overall effect of damping and delay circuits.
As indicated by equations (11.3.7)and (1 1.3.8)the firing instant is independent
of the delay angle a. However, the excursions of the firing instant must be limited
so that valves are not triggered outside a certain range dictated by hardware
limitations or operating requirements and to prevent loss of synchronism
between the firing system and the supply. Hence, limit stops are provided to
prevent the firing from occurring before a minimum delay (a,,,) of about 7
degrees or trigger the valve prematurely if a maximum delay (a,,,) of about 120
degrees occurs before Fi.
Valve extinctions are detected when their currents become negative. By linear
interpolation the actual zero-current crossings instants are determined and it is at
this instant that the valves are switched to the nonconducting state. Since it is not
possible to integrate through discontinuities, the integration steps must start with
the conditions prevailing immediately after the change in a valve state and these
are obtained by the interpolation process.
A valve extinction may produce a nonconducting condition in the converter,
i.e. the current path is interrupted due to there being no conducting valves in one
half of a bridge. Action is taken in such a case to ensure that the valves nearest to
the smoothing reactor are OFF and that the minimum number of valves remain
in the conducting state to provide a path to the common reference point, thus
avoiding the need for defining new network nodes.
Optimal extinction angle control (E.A.) is normally exercised with inverter
operation. This is achieved by comparing the smallest of the extinction angles per
cycle (y,,,) with a reference optimum ( y o ) and producing a correction given by

If the firing is retarded by the optimum control, the following commutation

angles will increase, hence the correction must be smaller in magnitude than the
error, i.e. K, must be smaller than unity. If the extinction angle of a particular
valve i goes below yo a safety control produces a firing advance

where K4 is a constant of similar value to K,, though normally chosen to be

greater to ensure safe operation.
 The inverter is provided with both E.A. and C.C. control. The value of the C.C.
setting being lower than for the rectifier. E.A. is the normal control and C.C. is
only called into action under transient conditions, when rectifier and inverter
currents may be different, or when the rectifier is unable to increase the link
current beyond the inverter C.C. setting. The inverter may also be provided with a
minimum delay angle limit which is usually greater than 90".

11.4. TRANSIENT GENERATOR MODEL

A phase-variable formulation has to be used in order to develop a general model
capable of handling symmetrical, unsymmetrical, nonlinear and distorted
waveform conditions with a single basic program.(3)As compared with the single-
phase 'd,q' model, the computational burden increases as a result of the time-
varying inductive coefficients.
For the purpose of sign conventions, in writing the differential equations the
machine is treated here as a motor, i.e. positive current flows into the positive
terminal of a circuit branch.
Direct and quadrature axes representation is retained for the rotor circuits as it
fits well into the actual geometry and winding arrangements. The position of the
rotor is specified with reference to phase a and the angle, 8, is measured from the
axis of phase a in the positive direction of rotation.
The following terminal voltage relationships can be written in matrix notation:

where :
w = dO/dt is the angular velocity

and
[ L ]= [y
]Lrs Lrr

- Lab- L,, w s 2(0 + 30) 1 - L, - La, cos 2(0+ 150)

-L,,cos~(~+~o)
---------------
1T--------------
- L,,COS~(O+ 150)
L, = L,+L,,cos~(~-I~o) -L,-L,,ws~(~-~o)
I
+L,cos~(~-I~o) - L , , C O S ~ ( O - 90)
---------------C--------------
- L, - ~ , , w s 2 ( e - 90) 1
I
L, + L , , c o s ~ (+~120)
- L,, cos 4(0 - 90) I
I
+ L,, W S ~ ( O+ 120)-

Lsr =
Torque matrix:

where

and

The model described contains two damper circuits aligned with the direct and
quadrature axes respectively. The machine inductance matrix is thus 6 x 6
square. The fourth harmonic of inductance variation, although not normally
known, is considered to be of relevance and has been included.
The AVR and governor representation when included, should be treated as a
separate set of equations. This strategy is recommended due to the wide variation
of the characteristic eigenvalues of such subsystems allowing a consequential
saving in computation time.
A detailed analysis of the electric network without generator control is
perfectly adequate for a number of cycles after a disturbance or for disturbances
lasting only for a few cycles, like a temporary valve mal-functioning of a stitic
converter. These are carried out over a few cycles of fundamental frequency only
and the generator controls (AVR and governor) can either be left out or
approximated by constrained algebraic relations. When specially required, both
the AVR and governor representations can be added to the present model as
separate subsystems taking the applied field voltage and the rotor angular speed
as the interface variables, or in the case of thyristor-controlled excitation it is
possible and desirable to represent the dynamic behaviour of the thyristor bridge.

11.5. TRANSMISSION LINES

The response of a transmission line to its terminal conditions can be analysed by
wave propagation techniques. The modal components used in the wave
equations require a sampling interval chosen so that travel times are integral
multiples of this sampling time. When the line forms part of an extensive power-
system such as shown in Fig. 11.9, with multiple a.c. and d.c. lines of varying
lengths, the travel times may be very different. The problem of different velocities
 A.C. lines
P

Fig. 11.9. H.v.d.c.-a.c. power system

of propagation, associated with the different modes of a multiconductor line,

compounds this so that a satisfactory sampling inverval which will coincide with
the converter switching operations will be too small for efficient computation.
A network approach is therefore adopted as described in Chapter 2 whereby
the line is segmented into an appropriate number of short line pi-segments,
according to the highest frequency of interest.
H.v.d.c. and h.v.a.c. multiconductor lines or cables are defined by their series
+
impedance (R jX) and shunt susceptance (B,) matrices per unit length. The pi-
segment elements are derived from the physical geometry of the conductors,
earthing and conductor-earth resistivities.
Defining the B matrix, whose order is the number of conductors, including
earth wires, and whose general element is given by:
Bij = loge(Dij/dij)
where
Dij = distance between the ith conductor and the image of the jth conductor;
dij= distance between the ith and jth conductors (i f j); or
dij= radius of ith conductor (i =j).
The capacitance matrix per unit length is C = 2n11~.B-~.
The useful part of C corresponds to the 'live' conductors since the earth wires
are at earth potential and may, therefore, be eliminated. This is achieved by
discarding the rows and columns in C corresponding to the earth wires.
The resistance and inductance matrices (series impedance) are given by:

where suffixes g, c and e signify, respectively, the effects of geometry, conductor

and earth path.
The contribution of the geometry to the total inductance is obtained directly
from the B matrix as L, = pB/2n.
The effect of the conductor on the resistance at low frequencies is limited to the
d.c. resistance per unit length and, where skin effect is significant, a corrected
power frequency value may be used. The internal inductance LCis constant and is
calculated using the concept of geometric mean radius. The earth return
contributions Re and Le can be calculated using an infinite series.
The earth wires cannot be entirely eliminated from the series impedance
matrices in the present study since the impedance matrix, given in terms of the
differential operator p as Z = R + Lp, has an inverse whose elements are,
generally, high degree polynomials in (p). It is only in special cases, when these
polynomials can be reduced, that the elimination of the earth wires may be
achieved without undue complication. Single frequency analysis, when p may be
replaced by jo,and lossless analysis are examples of such cases.
Therefore the equivalent circuit of a typical transmission line segment, and the
corresponding inductance and capacitance matrices, are shown in Fig. 11.10. The
elements of the capacitance matrix are related to the equivalent circuit by,
typically,

and

a' b' c'

a' coo Cob coc
b' cbo 'bb cbc
C' cco ccb ccc
(c

Fig. 1 1.10. Transmission line representation (typical pi-

section). (a) Equivalent pi-segment model; (b) Inductance
matrix; (c) Capacitance matrix
11.6. TRANSIENT TRANSFORMER MODEL
The transformer offers different impedances to the different components of
current depending on the type of magnetic circuit and on the connections of the
terminals and the neutral. Moreover, in the presence of static converters, phase
shifts due to different connections, must also be represented to cater for twelve-
pulse converter operation. It is, therefore, inaccurate to represent the transformer
by a simple series impedance.
Neglecting core losses, a single-phase transformer can be modelled in terms of
self and mutual inductances as:

where I , and I , are taken as positive when they flow into their respective
windings.
Numerically (in p.u.) the self inductances are equal, as are the mutuals. Their
relationship is approximately

where L,, is the total leakage inductance of the transformer. The value of the self-
term in modern transformers is very high (e.g. 1 percent magnetizing current) and
the leakage reactance is very small (e.g. 5 percent) so that in p.u.
L , , = L,, = 100 p.u.
and thus
L , , = L,, = 99.975 p.u.
The numerical value of the off-diagonal terms is therefore close to that of the
diagonal terms. Since the formulation requires inversion of the transformer
inductance, numerical instabilities could be created by the use of inaccurate data.
If there are further windings on the same core the matrix equation can be
extended to include them.
Three-phase transformers can be treated in the same way with the possible
simplification resulting from ignoring interphase mutuals and reducing the
equations to three independent sets for the three phases. This is accurate when
banks of single-phase units are used and acceptable with five-limbed units.
Effects like phase-shifts, zero-sequence circulating currents, neutral earthing
etc. are automatically catered for by the terminal connections. Off-nominal taps
can be represented by suitable modification of the self and mutual elements of the
tap-side winding, as described in Chapter 2, noting that the current in the tap-
winding is identical to that in the main winding and that the terminal voltage is
the algebraic sum of the main and tap-winding voltages.
If the converter transformers are subject to off-nominal tap change, it is
necessary to recalculate the elements of the inductance matrix. When the
nonlinear relationship between the derived equivalent circuit and the magnetiz-
ing characteristics are known, the elements of the coefficient matrix can be
continuously updated to correspond with the instantaneous values of the
currents, voltages or flux linkages. In such cases, the formulation using 6, and L,
suggested in Section 11.2 increases the programming complexity and it is easier to
use only the L, formulation.

Transformer saturation
In the coupled-circuit model used to represent the transformers, the magnetizing
current may be directly formed by summing the instantaneous primary and
secondary currents, with the appropriate sign. Multiplication by the magnetizing
reactance will result in the instantaneous magnetizing flux linkage, i.e.

where in per unit, the magnetizing reactance L, is equivalent to the self-reactance

,
(L, ,L,,) of the transformer.
A typical transformer magnetization curve(4) is indicated, with hysteresis
neglected, in Fig. ll.ll(a). As can be seen, the knee point of the curve is
well defined and the curve may be closely described by a two-slope linear
approximation as in Fig. ll.ll(b), since the slope in the saturated region above
the knee is almost linear, and the slope in the unsaturated region is the linear
unsaturated value of magnetizing reactance.
In cases where the instantaneous magnetization curve is not defined for a
particular transformer, it may be derived from the r.m.s. characteristics usually
supplied by the manufacturers, by a conversion technique, which ignores
hysteresis, eddy current losses and winding resistance^.'^)
When a transformer saturates, its operating point on the transformer
magnetization curve, which may be expressed as magnetizing flux linkage versus
magnetizing current in instantaneous values, is shifted into the nonlinear region.
In the dynamic analysis formulation, instantaneous magnetizing current is
directly available, whilst magnetizing flux is expressed by the time integral of the
voltage impressed at the transformer terminals. The state-space formulation,
however, requires constant coefficients over each integration time step, with no
discontinuities occurring in flux linkages.
The product of the instantaneous magnetizing current, obtained at time tj,
using equation (11.6.3),and the constant value for magnetizing reactance over the
previous interval (tj-, to tj), will provide an instantaneous magnetizing flux
linkage ($,). This value may then be located on the known magnetizing
characteristic to give an updated value for the magnetizing current (i,). The
following relationships apply with reference to Fig. 11.1l(b):
 Fig. 1 1 . 1 1 . (a) Typical transformer magnetization
characteristic; (b) Linearized transformer magneti-
zation characteristic

im = $1/L1+ ($m - $1)/L2 ( 1 1.6.7)

where L , and L2 are the gradients of the dual slope approximation.
To provide a generalized model for cases where the knee point is less well
defined, a multislope linearized approximation can be used, as an extension of the
two-slope approximation.
Since saturation will affect each phase independently, data for each phase of the
transformer bank must be stored separately. Reconstruction of the system matrix
L;' can be performed directly, due to its block diagonal structure, by simply
 '
inserting L, into the relevant location. If L, is dependent on L,, it must also be
reformulated.

11.7. STATIC ELEMENTS

Provision has to be made to represent various types of shunt-connected circuits
as shown in Fig. 11.12.

Fig. 1 1.12. Static shunt elements

Circuits (a) and (b) provide representation of local loads at the converter
terminals in order to investigate their damping effect and harmonic penetration.
Static capacitors (c) may be present for compensation and also to represent a
d.c. cable.
Circuits (4,(e) and (f) are typical of h.v.d.c. harmonic filters and line dampers.
Owing to their large-time constant, the converter harmonic filters have a major
influence on the converter waveforms following a disturbance and they have to be
accurately represented in the small-step dynamic simulation.
Static series elements consist of linear, uncoupled elements and are represented
by series inductances and capacitances. The latter are restricted, in that they must
be part of a capacitive network which provides a path to the common reference
point of the system; in practice, these capacitors are connected in series with
transmission lines and the restriction is met.

11.8. A.C. SYSTEM

Modelling of each component in an ax. power system comprising many
generators, transmission lines, etc., is possible, but for a large system this becomes
computationally uneconomic due to the size and complexity of the system.
Inevitably some assumptions must be made to obtain a less complex, yet
sufficiently accurate, model for the system. System components of particular
interest must be represented in detail; e.g. when a synchronous generator
dominates an a.c. system, the machine may be represented by its full transient
equations as in Section 11.4. With static converter plant, it is necessary to
explicitly model a.c. harmonic filters and the converter transformers. The
combined effects of other components may be treated by a form of network
reduction. Various problems occur on reduction of (parts of) an a.c. system to an
equivalent form. Not the least of these is establishing a model accurate for all
harmonics of the power frequency. Use of a simple fundamental frequency model,
in general, gives inaccurate and conservative results.
The a.c. system forms an oscillatory system with the a.c. harmonic filters which
may have one or more natural resonant frequencies of a low order. Disturbances
may excite these natural frequencies and subsequent waveform distortion, with
possible overvoltages, may occur in the a.c. system. These, in turn, will interact
with the converters' controllers and affect their responses.
Correct a.c. system representation on a reduced basis would require know-
ledge of the impedance-frequency locus of the a.c. network. From this infor-
mation, an a.c. system representation such as proposed by Hingorad6) and
illustrated in Fig. 11.13(a) may be used by the dynamic analysis program.
However, this information is subject to operational changes, such as line
switching, which can alter the shape of the locus quite drastically. As a result, the
availability of a locus for a particular system operating point is somewhat
suspect. To avoid this problem, some assumptions must be made which allow for
the use of a less complex model.

Converter

Fig. 11.13. (a) nth order network equivalent-to represent the har-
monic impedances of n resonant frequencies; (b) (Modified) Thevenin
equivalents of network
 Usually a locus would show a tendency for the system to be inductive for the
important lower harmonics (up to fifth). The a.c. filters will be capacitive below
the fifth harmonic and so a parallel resonant condition may exist around the third
or fourth harmonic, depending on the strength of the a.c. system. The stronger the
system the higher the resonant frequency will be in general. An equivalent circuit
which maintains a constant impedance angle over the lower frequency range and
provides realistic damping of harmonic voltages has been proposed by B~wles.'~
Giesner@)proposed an adapted version of this (Fig. 11.13(b))in which the source
impedance (which may be calculated from the short-circuit level at the converter
transformer busbars) is split so that there are two, not necessarily equal, reactive
elements with the resistive elements designed to give the required impedance
angle. The representation of a simple e.m.f. behind a series (source) impedance is
also provided for, although as explained, this will only be accurate at a single
frequency. The impedance angle varies markedly from system to system. Its
correct representation is important as its effect on damping of resonant
conditions such as overvoltages is critical. Typical values of 75 and 85 degrees for
receiving and sending end a.c. systems respectively have been ~uggested.'~) For
frequencies at or above the resonance of the high pass filter, the ax. system
impedance will be swamped. Thus the only assumption made is in the mid-range
of frequencies where the likelihood of a resonant condition in realistic strength
systems is small.
The use of an equivalent representation for an a.c. system therefore assumes
there are no network resonances yet gives an adequate treatment to possible a.c.
filter-system resonances. The possibility of associated over-voltages and wa-
veform distortion and their effect on the converter controllers is then catered for
adequately. Explicit busbar information would not be available at all busbars,
but the extremely efiicient simulation achieved in comparison with full a.c. system
representation justifies the use of such a model. However, for this equivalent to be
a realistic representation of the a.c. network when the study duration is long
and/or the disturbance severe it must include all the associated generator effects,
i.e. must be time-variant under disturbed conditions.

Time variant Thevenin equivalents

A dynamic analysis should accurately model the changing state of the a.c. system.
The use of a constant Thevenin equivalent throughout the full study period is not
realistic. The time response of generators throughout the network, must be
included, particularly when studying major disturbances, as machine effects will
then be considerable.
This requires a time-variant Thevenin equivalent for the network, which can
only be obtained from a simulation in time, since a single time solution, such as
performed by the fault study program (Chapter 8) can only provide equivalents
valid at that single instant of time.
The multimachine transient stability models described in Chapters 9 and 10
include a quasi-steady-state approximation of the d.c. link behaviour. This model
can be used to provide a Thevenin equivalent for each d.c. link terminal at each
integration step. This equivalent is thus a reduced representation of the a.c.
system as seen from the converter's a.c. busbar, and will include the associated
generator effects.
It will be shown later that interactive coordination between the Transient
Convertor Simulation and Transient Stability program can improve the
accuracy of the a.c. system dynamic equivalents.
In cases where the full a.c. system is represented, i.e. an equivalent repre-
sentation is not used, each machine may be represented by a time varying
equivalent, or in the case of a system-dominating machine, by its full transient
equations.

11.9. D.C. SYSTEM

The d.c. line parameters vary from scheme to scheme and the natural frequency of
oscillation ranges between 10 and 60 Hz in existing schemes.
In power-frequency related studies requiring mainly the behaviour of the d.c.
link at the two terminals and the fault point, it is sufficiently accurate to model the
line by a number of n equivalents, depending on distance, and including the
smoothing reactors. The behaviour of the smoothing reactors is reasonably linear
for currents up to two or three times the rated value and their presence limits
considerably the magnitude and rate of rise of current.
By way of illustration, Fig. 11.14 shows a monopole equivalent of the New
Zealand four-bridge bipole h.v.d.c. configuration described in Chapter 3 (Fig.
3.8). It includes a n-model for every 50 miles of overhead line and a shunt
capacitance representation of the sub-marine crossing. Such a model is con-
sidered sufficiently accurate for the purpose of analysing bipole d.c. faults and a.c.

LdRdrI r I r I r I r I r I

Data in per unit

Smoothing Reactors : Ld = 0.65807 Rd = 0.00042

S.I. overhead line : I = 0.09686 r = 0.00509 C, = C2/2
= 0.064752
N.I. overhead line : L = 0.040362 R = 0.00212 C3 = 0.026983

Submarine cable : C, =1.599743

Fig. 11.14. Monople equivalent for New Zealand h.d.v.c. link in d.c. line fault
studies
disturbances. Of course, the analysis of a single-pole fault on a bipole system
would require explicit representation of each pole and any mutual effects.

11.10. SOLUTION OF THE NETWORK EQUATIONS

Numerical integration
The choice of integration method is limited to the so-called single-step methods
which are self-starting. This is because multistep methods require information
store from previous steps and they are normally started by establishing a few
points using a single-step procedure. Taking into consideration the frequency of
the occurrence of discontinuities in converter bridges, it can be seen that a
multistep method would hardly start to operate before a discontinuity occurs and
the single-step procedure called upon again to restart the process.
Two methods for the solution of the differential equations namely, the fourth-
order Runge-Kutta method and the implicit inte'gration procedure based on a
trapezoidal approximation are explained in Appendix I.

Choice of state variables

By defining
$[ = L JI, $, = L ,Ik and Q, = C , V,
and using $,, $, Q, as the state variables, the magnitudes oft he derivatives are of
the same order as the voltages and currents of the network. The vectors I,, I, and
V, become dependent variables related to the state variables by algebraic
relations.
The relevant network equations developed in Section 11.2 are transformed to
the following (noting that E, and E, are zero):
Per unit system
Normal steady-state power-system analyses use a per unit (p.u.) representation of
voltages, currents and impedances in which these quantities are scaled to the
same relative order, thereby treating each quantity to the same degree of
accuracy. In dynamic analyses, instantaneous phase quantities are used in both
a.c. and d.c. networks and at the same time derivatives of relatively fast changing
variables are evaluated.
The usual cosinusoidal relation between a state variable and its derivative
indicates that the relative difference in magnitude is o , which may be high, so that
the degree of accuracy will not be the same for each. If the state variables are
changed by a factor o, (the fundamental a.c. system angular frequency), and
reactances and susceptances are used, instead of inductances and capacitances,
the integration process may be carried out as a function of the angular
displacement of a reference vector rotating at velocity o,, with the result that all
variables and coefficient matrices, and hence state variable derivatives, are of the
same relative orders of magnitude.
Therefore, by defining a base angular frequency o,, the values to be used for
the coefficient matrices are impedances and admittances based on this frequency,
for all parts of the network, including the d.c. link. The voltages and currents may
then be scaled to instantaneous phase quantities. Any convenient power-
invariant set of bases may be chosen for the network parameters. The integration
step will then be an angle rather than a time increment.

Solution of the transformed network equations

Equations (11.10.1)to (11.10.10)are solved for one integration step, starting from
given initial conditions and using the integration process specified.
Both integration methods used require the evaluation of equations (11.10.1)
to (11.10.10) several times per step. The Runge-Kutta method requires this
evaluation four times per step regardless of the step length. The optimal step for
the Runge-Kutta method will thus be the largest that gives a stable solution.
The implicit method requires the evaluation of the changes in dependent
variables and state variables during the iterative process of the implicit
integration algorithm. The smallest total number of evaluations of equations
(11.10.1) to (11.10.10) occur with a step length of about one degree for the
particular system under consideration.

Discontinuities
The valve-switching instants must be determined as accurately as possible, so
that the relevant network changes may be carried out.
Valve-firing discontinuities are decided beforehand by the control system and
the integration step-length can be changed so that the firing coincides with the
end of a step.
 Valve extinctions, on the other hand, can only be predicted at the expense of
slowing down the computation. It is preferred to detect extinctions after they
have occurred, when valve currents become negative. At this point, the time of
extinction can be assessed by linear interpolation. The time thus obtained is then
used to interpolate all the other quantities in the network. This is a first-order
approximation assuming that the variables change linearly over the step. If a
more accurate assessment is required then a backward integration to the
estimated time of discontinuity may be carried out with a further interpolation.
Voltage crossovers do not result in discontinuities but their occurrence must be
detected to provide a reference for the firing control system. This is achieved by
linear interpolation in a manner similar to the calculation of valve extinction
instants.

11.11. COMPUTER IMPLEMENTATION

Figure 11.15 shows the general flow diagram of a computer program developed
to implement the Transient Converter Simulation derived in previous sections. It
is important to separate, as far as possible, the processes indicated in the flow
diagram in order to simplify the program and its verification.

Input and establishing network equations

The data for a dynamic simulation study can be very considerable even for a
relatively simple system. For example, a system including a six-pulse d.c. link, a
single-circuit a.c. line, and the usual a.c. filters, can easily have upwards of 100
inductive branches, a dozen resistive branches and about 90 nodes of different
types. If the data had to be input in branch-list form, it would involve a great deal
of preparation and almost certainly some error, particularly since the inductive
and nodal matrices contain mutual elements which can increase the volume
considerably. Furthermore, any small change in the data, such as changing the
number of a.c. line segments, would involve a whole lot of data modification. A
more general approach should be adopted in which the data input is as simple as
possible and the job of expanding this data into the full network model is left to
the computer.
For each system component, a set of control parameters (composed of four
numbers) is read in, giving all the information required for expanding the given
component data and converting it to the form required by the program. As well as
these control parameters, a representative sample of the circuit constants of the
component is read in together with the one-line currents and other parameters,
such as control system constants and settings of converter bridges. This data is
systematically expanded into individual branches and nodes, the one-line
currents and voltages are converted into instantaneous phase quantities and
individual branch currents and node voltages are calculated and assigned. The
new network nodes, created by the expansion of the data, are assigned a type in
accordance with the types of branches connected to them as defined in Section
 I lnput data and establish equations I
I
lnput fault data
&

I Monitor and control converter operation:

any switching ? I
NO produce output

Time variant inductances ? i.e. Lf = f ( l ) ?

e.g. detailed synchronous machine models
1 $ YES

relevant blocks in L f .
Recalculate L,-'

Is Ly time variant?

Form L y

I Solve network equations for one step I

.t.
Advance time by h
1
1

I Time for output? 1

write output to

Time for disturbance ?

+YES
r ~ o d i f yequations according
NO to fault data. Input any
further fault data
v I
End of study?
NO

state information on

Fig. 1 1.1 5. Flow diagram of digital computer program control

11.1. A set of cross-referencinginformation is created relating the system busbars

to the new network nodes, ordering the different types of nodes into their
respective sequences and relating them to the network nodes. If the component is
a converter, bridge valves are set to their conducting states from knowledge ofthe
a.c. busbar voltages, the type of converter transformer connection and the initial
delay angle which is given as part of the data.
The procedure described above, when repeated for all the components of the
system, generates the following:
(i) L,, R,, L,, R,, C,; in compact form, together with their corresponding
indexing information ;
(ii) the branch lists for inductive and resistive branches in terms of the set of

t
Read progmm control data
I
).
I System already established? 1

Type o f component
End Line Transformer Synchronous Filter Converter Reduced
machine system
0 1 2 3 4 5 6

I I I
Follow relevant procedure to: read control and
circuit porameters,establish new nodes and
branches, assign node types and sequences, set
initial currents and voltages, set up primitive
coefficient matrices,set up branch lists, etc.

Form constant incidence matrices,

and I$ and LI
*
$.
Form initial L a and Kak

Fig. 11.16. Flow diagram of data input and initial value establishment
 new network nodes created by the procedure, together with the type
designation and cross-referencing information;
(iii) the initial values of branch current and network node voltages;
(iv) for converter bridges, the basic (noncommutating) nodal solution matrix
L, and its indexing information, and the initial valve states in each bridge.
Alternatively, network data and initial conditions, stored from a previous
(initializing) run may be input in its final expanded form.
The network equations are established by defining all the coefficient values of
the variables $,, $, and Q,. This process involves the inversion of the matrices L,,
L,, R, and C,, the formation of permanent incidence matrices, including an
auxiliary connection matrix used in the calculation of valve currents, and the
formation of Rg, L,, L,, and K,,, i.e. all the variable matrices. All dependent
variables and the derivatives of state variables are then calculated so that the
solution of the network equations may commence.
Figure 11.16 shows a general flow diagram of the process.

Monitoring and control of converter bridges

At each step of the integration process, converter bridge valves are tested for
extinction, voltage crossover, and conditions for firing. If indicated, changes in
valve states are made and the control system is activated to adjust the phase of
firing. Moreover, when a valve switching takes place, the network equations and
the connection matrix K,, are modified. The flow diagram in Fig. 11.17 illustrates
the monitoring and control procedure.

External disturbances
Externally applied disturbances, as opposed to those inherent in the network as a
result of valve switchings, can be identified by a set of data defining the
disturbance and indicating the time of application. Besides the application of a.c.
and d.c. faults other typical requirements for the dynamic simulation are:
(i) a.c. fault clearance;
(ii) d.c. line charge/restart/reversal;
(iii) if a d.c. fault is to occur, detection parameters need to be established.
Simultaneous application of disturbances should be possible.

Output
The main output stream dumps the calculated results onto a mass storage
medium at specified intervals, say 10-20 degrees of fundamental frequency.
However, output immediately before and after switching operations is also
necessary. A separate program is recommended to retrieve the dumped results
and produce time-response graphs of the required variables. It is also advisableto
produce graphs of variables which are not explicitly defined in the program, such
 t
I Calculate valve currents 1

For each bridge: if extinctions

detected determine earliest extinction
and change valve state

J.
IExtinction detected 71
YES
Interpolate all state
variables back to extinction
~nstant.Calculate
dependent variables

t
For each bridge
check for cross - overs,
:

and activate control

system in relevant
bridge if they occur

firing due, issue firing

( Any switching opemtion above? I

NO +YES
Produce output
modify L8 and KBk
set next

T
If time to next firing
+
output time

less than integration

step, reduce integration
step.

Fig. 11.17. Flow diagram of monitoring and

control of h.d.v.c. converters

as commutation voltages. Finally, it should be possible to magnify selected parts

of the plottings to show more detail.

Interactive coordination with transient stability analysis

The role of a multimachine transient stability program in obtaining time-variant
a.c. system representation has been discussed in Section 11.8.
The Transient Converter Simulation (TCS) results need to be processed to
provide information on fundamental power and voltage at the converter
terminals for each integration step of the transient stability study.
In this way, the quasi-steady-state approximation to d.c. link performance can
be replaced by a more accurate representation of the link performance and
further iterations can be made if necessary to improve the interaction between the
two programs as detailed in Fig. 11.18. To prevent iterating between the two

Dynamic analysis: performed inn

v
If observation of dynamic
analysis indicates mismatch with
quasi-steady state d.c. link
model intmnsient stability
study adjust quasi -steady state
perf0;mance occordin ly New
tran~ientstability will?hsn
prov~demore accurate Thevenin
equivalent sequences

FStage i

IStagen -end of dynamic amlysis I

1
Process dynamic analysis
information toenable use in

Transient stability study

using processed dynamic
analys~smformat~ony~eldsnew
Thevenin equivalents

YES Check difference between the

q u e n c e of equivalents-more
~terattonsnecessary ?

Fig. 11.18. Flow diagram o f iterative interation between dynamic

analysis and transient stability computer programs
programs, an expensive exercise (indicated by the LHS loop in Fig. 11.18), the
following alternative approach is suggested.
It is possible to observe the trends as the TCS progresses through consecutive
stages, and if necessary, to interactively modify the performance of the quasi-
steady-state d.c. link model over the period already dynamically simulated, so as
to match more closely the dynamic simulation results. This is shown by the RHS
loop connected between stages (i-1) and (i) in Fig. 11.18. An updated sequence of
time-variant network equivalents are then provided which can be used from stage
(i) on. Although several short duration transient stability runs may be required
with this approach, these are less expensive than the alternative iteration scheme.

Incorporation of time variant network equivalents(g'.(lO)

In the presence of multibridge h.v.d.c. converters, the transient converter
simulation requires a very small integration step (of the order of 0.1 ms). On the
other hand, the transient stability program, used to determine the generators'
response during disturbances, usually provides reliable information with in-
tegration steps of 10 ms, i.e. one for every 100 dynamic simulation steps.
In the transient stability program, the gradual changes of a.c. system
parameters from the subtransient to the transient, as obtained from the generator
response, are used to update the injected currents at the terminals of each
generator. The updated current injections yield a new Thevenin source voltage,
while the Thevenin impedance remains constant.
The r.m.s. magnitude (E) and phase angle (0) of each Thevenin voltage at times
,
tj and tj+ as calculated by the transient stability study are stored in a data file.
The rates of change of these Thevenin voltages over each time step of the transient
stability study are sufficiently similar to allow a linear interpolation to find an
appropriate equivalent at each time step of the transient converter simulation, i.e.
at time t,

where

and

The dynamic analysis then calculates instantaneous phase values for each
Thevenin equivalent source voltage at t, from these quantities, by
e, = Ex cos (8,+ wt)
eb = Excos (0, + wt - 120')
 7
Monitor and control converter operation:
any switching ?

Modify equations,
produce output

I ? i.e. L1 = f ( f ) e.g.
Time variant ~~..'.rctances
saturation or synchronous machine models I
Calculate new values for
relevant blocks in L!
recalculate ~ 1 - I

Any time variant network equivalents

data produced by transient

Solve network equotions for one step

b

+
Advance time by h

Fig. 11.19. Modification to program control flow diagram

to accommodate time variant Thevenin equivalents and
transformer saturation

The necessary modifications to the overall program control, illustrated in Fig.

11.15, to incorporate a facility for time variant network equivalents, are
illustrated in Fig. 11.19.
Careful alignment of the transient stability data file with the dynamic analysis
is necessary to correctly identify fault application and removal times. An initial
solution at t = 0 by the transient stability program which yields information in
the form of a load flow, provides the initial Thevenin equivalents. To shorten the
establishment of prefault initial conditions, the reference angle in the a.c.
system@)is adjusted so that the run begins within a commutation-free period.
Thus, all future angles obtained from the transient stability data file must be
similarly adjusted.
The dynamic initial conditions are obtained using the constant initial
Thevenin equivalents. Fault application initiates the process of obtaining the
time variant Thevenin equivalents from the transient stability data file, as
described above.
A consequence of using a time-variant equivalent is that frequency deviations
occur as the a.c. system(s) swing. Provided control information in the form of I,,
dI,/dt (both nonzero) and di (from crossovers) is still available, this will not cause
any problems. However a must be measured from the crossover of voltages (as for
d), if a correct indication of the absolute value of firing angle is required.
Alternatively, a reference to the system frequency may be formed so that
distortion of the a.c. voltage waveforms does not confuse the firing control
pattern when viewed as an absolute firing angle (a). The system swing is then used
to modify the interval between firings (AFi), which are 60" in the steady state.
Under disturbed conditions the interval will be (60" + APi). To compensate for
frequency shift the interval will become

where At3 = Oj+, - Oj is the change in Thevenin source angle over the relevant
transient stability step. This may also be used to yield a value of a relative to the
actual a.c. voltage waveforms.

11.12. PROGRAM OPTIMIZATION AND USE

Scope
The computer program must handle a combined a.c.-h.v.d.c. system of
reasonable size and flexible configuration. Hence it must be capable of referring
to a h.v.d.c. converter not only in the special context of a rectifier-line-inverter
arrangement, but also in terms of a multi-terminal h.v.d.c. system, for example.
The number of transmission lines, transformers, etc., are limited only by the
storage allocated to the combined total of their equivalent circuit elements.
Provision should be made for different transmission line configurations, for both
a.c. and d.c. operation, and the number of segments used to represent the lines
should be variable within the limits of the total storage available.

Sparsity storage
The coefficient and incidence matrices in equations (11.10.1) to (11.10.10) contain
a large proportion of zero elements (e.g. Fig. 11.20). Only the nonzero elements
need be stored in order to save computer storage and time in performing matrix
operations. Further reduction in storage is achieved by storing only a sample of
any constant sets which are repeated, i.e. assuming linearity and balance, all
three-phase matrix elements, e.g. multiple-segment transmission line series
inductance matrix, and each phase of a bank of single-phase transformers. This
will not apply to transmission line nodal capacitances nor to resistances, the
latter (R,, R,, R,, RB) are stored as simple diagonal vectors whose indexing
information is inherently given by the relative position of the elements.
 Branch No 1 2 3 4 5 6 7 8 9 10 11 12
From 1 4 2 5 3 6 19210311
To 0 7 0 7 0 7 8 1 1 8 9 8 1 0

(e)
Fig. 11.20. Storage of coefficient and incidence matrices. (a) Equivalent
circuit; (b) Full inductance matrix; (c) Compact inductance matrix;
(d) Branch list; (e) Node-branch incidence matrix

Coefficient matrices
The representative samples of nonzero matrix elements can be stored in a vector
of real constants C, and the indexing information in three integer vectors LC,
KC, LK, as in the example in Fig. 11.19(c).C contains the nonzero elements of
the full matrix, scanned row-wise, ignoring repeated sets. The vector LC gives the
relative address in C of the start of information of each row that is defined in the
full matrix. KC contains the column numbers of all nonzero elements in the full
matrix while LK gives the relative address in KC of the start of information for
each row.
 The branch and node coefficient matrices (L,, L,, L,, C,) are also ordered in
such a way that they are composed of block-diagonal submatrices of low order.
Each block is a square matrix representing a set of mutually coupled branches
with few, if any, null elements. The full matrix inverses are therefore identical to
the original matrix in structure. This makes it relatively easy to calculated inverse
and matrix-by-vector products in a systematic manner.

Incidence matrices
Incidence matrix information can be stored in two integer vectors: the coefficient
vector (K) and the associated index vector (L).
Taking a node-branch incidence matrix as an example, L gives the relative
address of the start of information in K for each node. The coefficient vector K
contains the numbers of branches that are connected to each node, preceded by a
minus sign if the node happens to be the receiving-end of the branch in question.
This is effectively storing the matrix by rows, replacing a + 1 by its column
number, a - 1 by the negative of the column number, and completely ignoring
zeros. Figure 11.20(e) shows the node-branch incidence matrix for the example
given.

Implementation
It is recommended to use subroutines to perform the tasks of forming, storing and
indexing the coefficient and incidence matrices. As well as these tasks, which are
basically required at input, the solution of the network equations is best achieved
by using special routines to perform the coefficient matrix inversion, the
evaluation of matrix-by-vector products, etc.
The product of a compactly-stored incidence matrix D and a vector V can be
obtained by algebraically adding the elements of V whose relative addresses and
signs are given by the elements of K pertaining to each row in D. The product of
the transpose of D and a vector can be obtained by a similar modified procedure,
and it is therefore only necessary to store one form of incidence matrices. The real
coefficient matrix-by-vector product can be achieved by summing up the
products of the nonzero elements in each row and the elements of the vector
whose relative addresses are given by the column numbers stored in KC.
Multiple products, such as those used in equations (11.10.5) and (11.10.6),
can be formed by scanning the expressions from right to left, progressively
carrying out single matrix-by-vector multiplications until the final product is
reached.

11.13. GENERAL CONCLUSIONS

A general method has been developed for the computation of a.c.-d.c. system
disturbances which combines the efficiency of transient stability modelling and
the accuracy of transient simulation of converter behaviour.
The transient converter simulation uses time-variant Thevenin equivalents,
formed by linear interpolation, from the discrete values obtained at every step of
the transient stability study. Similarly, transient stability compatible equivalents
are formed from the transient converter simulation during the disturbance. Such
interaction permits the use of reduced equivalent circuits and results in moderate
computing requirements. Every cycle of real-time simulation requires about 14
seconds of CPU time on an IBM-3033 computer. As an indication of computing
cost, a complete run of a typical disturbance required 3 minutes of CPU time.
Moreover, the transient converter simulation is only required for short runs, i.e.
during and immediately after the fault period. The extra computing time required
for stability studies is relatively insignificant.
The use of a mass storage medium during simulations allows postprocessing of
results. Any variable or combination of variables can be plotted on any scale
desired, without the transient converter simulation being re-run.
The interactive model developed provides a useful tool in assessing converter
controllability during disturbances and the effect of alternative control and
protection schemes. It also provides accurate information of fault current and
voltage waveforms in the vicinity of large converter plant, the subject of Chapter
12. Such information is essential for the rating of switchgear and the setting of
protective relays.
Finally the accuracy of the model used to simulate the disturbance is critical in
the assessment of transient stability, a subject discussed in Chapter 13.

11.14. REFERENCES
1. J. Arrillaga, H. J. Al-Khashali and J. G. Campos Barros, 1977. 'General formulation
for dynamic studies in power systems including static converters', Proc. ZEE, 124
(1 1), 1047-1052.
2. J. D. Ainsworth, 1968. 'The phase-locked oscillator-A new control system for
controlled static converters', Trans. ZEEE, PASS7 (3), 859-865.
3. J. Arrillaga, J. G. Campos Barros and H. J. Al-Khashali, 1978. 'Dynamic modelling of
single generators connected to h.v.d.c. converters', Trans. ZEEE, PAS97 (4), 1018-
1029.
4. H. W. Dommel, 1971. 'Nonlinear and time varying elements in digital simulation of
electromagnetic transients', Trans. ZE E E, PAS-90, 2561-2567.
5. H. W. Dommel, 1975. 'Transformer models in the simulation of electromagnetic
transients.' Paper 3-114, 5th Power System Computation Conference (PSCC),
Cambridge, England.
6. N. G. Hingorani and M. P. Burberry, 1970. 'Simulation of a.c. system impedance in
h.v.d.c. system studies', Trans. ZE E E, PASS9 (5/6), 820-828.
7. J. P. Bowles, 1970. 'A.c. system and transformer representation for h.v.d.c. trans-
mission studies', Trans. Z EE E, PAS-89 (7), 1603- 1609.
8. D. B. Giesner and J. Arrillaga, 1971. 'Behaviour of h.v.d.c. links under balanced a.c.
fault conditions', Proc. ZE E, 118 (3/4), 59 1-599.
9. M. D. Heffernan, K. S. Turner, J. Arrillaga and C. P. Arnold, 1981. 'Computer
modelling of a.c.-d.c. disturbances-Part I. Interactive Coordination of generator
and converter transient models', Trans. ZEEE. Winter Power Meeting, Atlanta.
10. K. S. Turner, M. D. Heffernan, C. P. Arnold and J. Arrillaga, 1981. 'Computer
modelling of a.c.4.c. disturbances- Part 11. Derivation of power frequency variables
from converter transient response', Trans, ZEEE. Winter Power Meeting, Atlanta.
A.C.-D.C. System Disturbances

Conventional faulted system analysis, as described in Chapter 8, cannot represent

the highly nonlinear response of h.v.d.c. converters to fault conditions. Such
response and subsequent effects are often neglected without proper justification
and fault studies are carried out by ignoring the presence of the d.c. link or
representing it by quasi-steady-state models.
While this simplification may be justifiable in particular cases, the power
ratings of h.v.d.c. links are generally too large to be ignored and their speed of
controllability has a profound effect on the entire system behaviour during and
immediately after the fault. There is, therefore, a need for detailed transient
converter simulation with realistic representation of both the a.c. and d.c. systems
as explained in Chapter 11. According to their origin, a.c.-d.c. disturbances can
be divided into converter faults, a.c. faults and d.c. faults.
Converter faults, though relatively frequent, are normally self-recovering and
have little influence on the system as a whole. Moreover, from the modelling point
of view, the converter operating modes resulting from internal converter
disturbances are not different from those produced by converter faults con-
sequential to the occurrence of external disturbances.
This chapter describes the application of the dynamic simulation program
developed in Chapter 11 to the analysis of a.c. and d.c. fault conditions. The New
Zealand a.c.-d.c. power system is used as a test model for the results.

A.C. FAULT MODELLING

12.2. REDUCED NETWORK REPRESENTATION
To determine an adequate equivalent network, the areas of specific interest must
first be identified. For a.c. fault studies, the most important points of interest for
the observation of overcurrents, overvoltages and waveform distortion will be the
converter terminals and the fault locatidn. Therefore, in general, explicit
information will be required at three locations, viz. at the a.c. terminals of each
converter, and the fault location.
 The New Zealand Primary Transmission system consists of two separate 220-
kV a.c. networks of comparable capacity, containing 21 buses in the North
+
Island, and 19 buses in the South Island, which are interconnected by a 250-kV
h.v.d.c. transmission link. As the system is not dominated by any particular
component (other than the d.c. link), it is appropriate to consider an equivalent
network representation, with just the three locations detailed above being
modelled explicitly. Effectively then, at each node of interest a Thevenin
Equivalent must be obtained for the a.c. system, as seen looking into the system
from the relevant node.
Approximate equivalent networks can be obtained from an a.c.-d.c. fault study
programme with quasi-steady-state representation of the d.c. link.'') The
program uses nodal system analysis, and the steady-state prefault nodal voltages
of the a.c.-d.c. system are described by the following matrix equation.

where
Z is the bus impedance matrix,
1 is a vector of nodal injected currents due to a.c. generators,
and
J is a vector of nodal currents injected by the d.c. converters.
The model of the d.c. link assumes that the converters retain controllability
during the disturbances, and thus maintain the current setting. When a three-
phase fault occurs in one of the a.c. systems, represented by a current If flowing
out of the faulted bus p, the quasi-steady-state model resolves the d.c. link
equations. With the d.c. link on constant-current control (at one of the two
converter stations), the magnitude of the current injections, J f , will remain fixed,
but the respective phase angles of the injections will depend on the terminal
voltages and the d.c. link operating conditions (a,, 4).
From the matrix equation (12.2.1), the following expressions are extracted for
the nodal voltages at the converters and fault buses (Fig. 12.1).

L.-.-.-.-.-.-.-.-.-.-.-i.
Fig. 12.1. Reduced equivalent network
 vi= vm+ z m m ( lfi Ji)+ Zmn(lf+ J f ) + ZmpIf (12.2.2)
Vf = Vn + Znm(li+ Ji)+ Znn(I;C+ J:) + z,,I~ (12.2.3)
i' = - zfIf = vp + zpm(I/,+ J/,) + zpn(I{+ J{) + zppIf (12.2.4)
where
Zf is the fault impedance,
If is the fault current, and
Vm,Vn,Vp are the prefault voltages at the rectifier, inverter and fault buses
respectively, in the absence of d.c. transmission.
Equations (12.2.2) to (12.2.4)describe a matrix set containing three interrelated
Thevenin equivalents. Each equation contains two terms which couple that
equivalent to the other two equivalents, viz.
Z, j(If + J!) for k~ m, n, p

where
Ji = 0 and I; = If
This equation set can be easily incorporated into the transient converter
simulation of Chapter 11 by modifying equations (11.lo. l), (11.lOS) and (11.10.8)
to allow for the mutual coupling term, which has both real and imaginary
components of impedance, i.e.
(i) R, now contains off diagonal terms, and
(ii) L, contains additional off diagonal terms.
The initial conditions are obtained using the process described in Section
11.11. The quasi-steady-state fault study, which provides the Thevenin equiva-
lents, also provides the steady-state prefault data necessary for the dynamic
analysis to begin the establishment of correct dynamic initial conditions.
The applicability of the equivalent circuit at harmonic frequencies has been

Fig. 12.2. Equivalent h.v.d.c. representation for a.c. faults

discussed in Section 11.8. The test results included here relate to a purely
fundamental frequency a.c. network impedance. However, whenever the
impedance-frequency locus of the system is known it can be modelled in the
program by appropriate equivalent networks as illustrated in Fig. 11.13(a).
As explained in Section 11.9 for a.c. fault studies, the d.c. line representation
may be simplified to a double-bridge monopole equivalent, with three n-sections
for the South Island overhead line, a large shunt capacitance for the submarine
cable and a series impedance for the short overhead line in the North Island. The
latter is incorporated into the smoothing reactor data and the complete d.c. side
representation is shown in Fig. 12.2.

12.3. FAULT APPLICATION AND REMOVAL

The following typical sequence is used to model the behaviour of the a.c. fault.
(i) detection of fault with relevant discrimination;
(ii) operation of the relevant circuit breakers to isolate the fault within 3-10
cycles;
(iii) after a period sufficient for the fault path to deionize, the breaker is
reclosed thus completing the original circuit;
(iv) if the fault re-establishes, the sequence is repeated for up to two to three
times and then if the fault still persists, the breakers will permanently
isolate the fault.
The node specification at the fault location is determined by the degree of a.c.
system representation, i.e. equivalent circuit or full representation.

Equivalent a.c. system representation

A fault branch (Zf), which may be resistive or inductive, is established in the
original branch list. Any value of Zf which is not infinite in the prefault period
would create an incorrect system loading. However, very large values of Zf affect
the rate of convergence of the numerical integration of the state variable
equations.
This can be avoided by including in the Thevenin equivalent, as seen from the
fault point, a generating source exactly equal to the effective loading created by
the noninfinite shunt impedance to ground at the fault point. By the principle of
superposition, the dynamic model will then see the fault bus as an open-circuited
terminal which will have the correct voltage.
At fault application time, the shunt impedance is altered to form the
appropriate faulted network.
In the case of a remote fault with a purely resistive fault branch of zero fault
'
resistance, R; will be infinite and Rg will be zero. This would necessitate a very
short integration step to achieve a reasonable rate of convergence for the
numerical integration process. Instead, a resistive-inductive fault is selected to
avoid convergence problems. In the case of a converter terminal fault (with high-
pass filters connected), either a resistive (4)or inductive (jX,) fault of any
magnitude can be simulated.
With a three phase-to-earth fault, each phase will have an equivalent which is
grounded through a fault branch. For a single-phase fault the two healthy phases
will retain their prefault configurations. Similarly, a phase-phase fault can be
simulated by connecting the relevant phases together and if the fault involves
earth they may be connected to ground as well.
The actual time of fault clearance is determined by the following sequence of
events. At a prescribed time following fault application, for example, five cycles, a

NO
simulation i.e. Fault clearance
4 YES

fault branch:
1
For each node of
fault location
Fault already cleared apply fault
on this phase

Much greater perform extrapolation

to estimate h' -remove fault

_I

Ia-node fault location -node fault location

Alter connection matrix Ks,

or Kg/ ar elemental value 4
or remove resistance R, I
I Step length= min ( h , ht) I

Fig. 12.3. Flow diagram of actions taken for fault application/removal

flag is set whjfiindicates that fault clearance is to be initiated on the relevant fault
branch(esj 'at the next zero crossing of current in the fault branch(es).
Extrapolation is used to predict the time when a zero crossing will occur; i.e. if tj is
the present time step, and h the step length, then
tj=tj-l + h (12.3.1)
Therefore, if the current in the fault branch at time tj is Ij, then

so that if h' is less than the next integration step (the nominal value as calculated
in the integration routine), the next time step will be

which, assuming linear extrapolation, will coincide with the zero crossing of fault
current. Fault clearance will then be effected as described above by alteration of
an elemental impedance value to re-establish the prefault circuit.
This process is repeated for each of the phases which are faulted. A flow
diagram depicting the actions taken for fault application and removal is
illustrated in Fig. 12.3.

Full a.c. system representation

An example of fault application and removal with full a.c. system representation
for a (three-phase-to-earth) a.c. line fault is now described with reference to Fig.
12.4(a):
(i) Insertion of the relevant fault impedance at the fault point in parallel with
the shunt capacitance of the n model corresponding to that location, by
alteration of a pre-established connection matrix element.
(ii) Subsequent to the specified circuit-breaker operating time, the line is
isolated by reforming the connection matrix elements associated with the
line ends (breaker locations), on detection of current zeros there.

Fig. 12.4. (a) Full system representation of a line fault and circuit breakers; (b) Inclusion
of circuit breaker resistances
 (iii) After a specified time, the breakers are reclosed and the fault to ground is
removed by reforming the original connection matrices.
Each of the breaker operations can simulate the inclusion of resistance in the
makebreak sequences. A series resistive element may be included at the relevant
line end (r,, r, in Fig. 12.4(b)),the value of which may be practically zero without
affecting convergence patterns, since the line ends are generally connected to
other lines, and are therefore in general a nodes.
Opening of the circuit breaker thus involves
(i) increasing r, and r, at a rate defined by the breaker contact resistance us
time behaviour ;
(ii) at a certain resistance, setting elements a,r, and a,+,r, to zero in the
connection matrix K,,.
Reclosing the circuit breaker involves:
(i) removing Zf by setting element a, Zfto zero in K,,;
,
(ii) reinstating elements a,r, and a, + r, to their original values (1, - 1)in K,,;
and
(iii) decreasing r,, r, to comply with the preinsertion resistor values.
In cases where breaker contact resistance is not modelled, only step (ii) in the
opening action is simulated. Similarly in cases where no preinsertion resistance is
used, only steps (i) and (ii) are performed, with r, and r, equal to zero.

12.4. TEST SYSTEM AND RESULTS

The test studies illustrated are primarily concerned with the fault and immediate
postfault periods. Extensions of the study-periods to investigate system stability
are discussed in Chapter 13.
As an example of a severe disturbance, the case of a three-phase fault at the
inverter a.c. busbar has been selected. Some fault impedance is included to retain
sufficient terminal voltage for the commutations (approximately 10 percentage of
nominal).
The fault current, illustrated in Fig. 12.5(a), apart from the initial d.c. offset,
clearly demonstrates the effect of the time-variant a.c. system equivalent voltage
obtained from the transient stability study as explained in Sections 11.8 and
11.11. From an initial value of 1 p.u. the Thevenin source at the inverter end
reduces to 0.5 p.u. after five cycles. This is depicted in Fig. 12.5(b). The phase
angle at the inverter terminal is shown in Fig. 12.5(c)and the combination of Figs.
12.5(b) and (c) also indicate the variation of the equivalent source at the rectifier
end.
The corresponding a.c. voltage waveforms at the inverter terminal are
illustrated in Fig. 12.5(d). Following the behaviour of the source voltage in the
postfault period (Fig. 12.5(b)), the terminal voltage levels show a gradual
recovery.
 0.0

-125.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 10 80 9.0 10.0 11.0 120
Time (cycles)
(a)

25
Rectifier

0.0
Inverter

0.01 I -25.01 , , , , , 1
0.0 2.0 4.0 6.0 8.0 10.0 12.0 0.0 2.0 4.0 6.0 80 10.0 12.0
Time [cycles ) Time ( cycles )
(c)

-1.5
0.0 1.0 2.0 3.0 4.0 50 6.0 7.0 8.0 9.0 10.0 11.0 12.0
Time (cycles
(d)

Fig. 12.5. Typical results for a three-phase short circuit at the inverter end. (a) A.c. fault currents; (b)
Thevenin voltages; (c) Thevenin angles; (d) Inverter a.c. voltages

The behaviour of the d.c. link is indicated in Fig. 12.6, which shows the d.c.
powers at the rectifier and inverter ends.
A second example, involving a less severe disturbance, is the case of a three-
phase fault sufficiently distant from the rectifier end to maintain about 65 percent
of the nominal voltage at the rectifier terminal. Results for the a.c. current and
 0.8

0.0

-0.7

-1.5
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0
Time (cycles

Fig. 12.6. Converter powers for a three-phase short circuit at the inverter end

-250.01
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11 0 12.0
Time (cycles

(a)

Time ( cycles)

Fig. 12.7. Typical a.c. results for a threephase short circuit at the rectifier end. (a) A.c. fault
currents; (b) Rectifier a.c. voltages
 0.01
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0
Time (cycles)

Fig. 12.8. Converter powers for a three-phase short circuit at the rectifier end

voltage waveforms are illustrated in Figs. 12.7. The variation of d.c. powers is
plotted in Fig. 12.8.

D.C. FAULT MODELLING

12.5. OVERCURRENT CHARACTERISTICS O F D.C. LINE FAULTS
In the absence of circuit breakers, d.c. transmission faults can not be cleared by
temporary switching followed by autoreclosing. On the other hand, the inherent
C.C. control of h.v.d.c. converters prevents large and sustained fault currents.
The levels of rectifier and fault overcurrents resulting from a d.c. line short
circuit depend on system and fault parameters, fault timing and location, mode of
converter operation, and subsequent control action. Peterson'') investigated the
level of transient rectifier overcurrent by isolating several parameters to
determine their effects. His analysis was based on a quasi-steady-state d.c.
representation, and ignored the effects of the a.c. system.
However, the assumption of a sinusoidal voltage beyond the commutation
reactance is not generally valid. The increased rectifier current during the fault
period will produce harmonic current imbalance in the a.c. system, which will
always result in a.c. waveform distortion for an a.c. system with finite short-circuit
capacity. Moreover, the extra alternating current required as I, increases must be
supplied from the a.c. network, thus causing further voltage regulation at the
converter terminals. The use of an infinite a.c. system representation will in
general yield pessimistic levels of rectifier overcurrent.
The presence of the smoothing reactor limits the magnitude and rate of rise
of overcurrents. Under high overcurrent, reactor saturation would increase
substantially the peak fault current and a special design of reactors is required to
retain their nominal inductance with up to two to three times rated current. The
effect of resistances in the converter transformers and smoothing reactors on
peak currents is small.
Since the d.c. system contains only passive elements, the main source of fault
current is the a.c. system, with d.c. line discharge contributing additional fault
current. The maximum fault current magnitude is determined by the impedance
between a theoretical a.c. infinite bus and the fault location. Therefore, strong
rectifier a.c. systems will result in higher fault and rectifier overcurrents than
would a d.c. line fault with a weak (i.e. low S.C.R.) rectifier a.c. system. A zero
resistance fault on the line side of the rectifier smoothing reactor is therefore the
most severe in terms of location.
The initial overcurrent is limited by the surge impedance of the d.c. line (which
includes the line capacitance, inductance and smoothing inductance). Since
control action is only possible at discrete intervals (every 60" for six-pulse, and 30"
for twelve-pulse operation), even with instantaneous detection, a transient
overcurrent will occur. However, control action, initiated by a protection scheme,
which retards the firing into inversion, will limit the maximum overcurrent level
when compared to normal regulator action (C.C.),since the latter has time delays
(circuit time constants) and/or gain limits incorporated in their design to ensure
stable operation in the steady state.
Severe oscillations of d.c. voltage and current can occur following normal
control action, due to possible resonance effects between the d.c. line and the
control system.(3)This highlights the need for an overriding protective action to
limit overcurrent stresses and ensure rapid fault clearance and recovery.
Normally, no more than one commutation occurs in twelve-pulse operation after
a line fault, before protective action is taken by the current control system.
Rectifiers operating with minimum firing angles at fault inception will exhibit
higher overcurrents than those with larger firing angles, since the effective voltage
driving the fault is then greater. Figure 12.9(a) indicates the effect of the rectifier
delay angle on the maximum level of rectifier overcurrent, at the time of the fault
inception. These curves assume instantaneous fault detection, and the fault is
applied immediately after a valve firing so that no control action can occur until
the next firing instant. The levels are pessimistic in that they are obtained with an
infinite a.c. source,(2)so that for six-pulse operation:

Ipx= (2 + cos a, - f i sin ar)rr/6Xc (12.6.1)

and for twelve-pulse operation:
lrx= (fi+ f i - (2 + 3)
sin + c o s a , ) r r / l 2 ~ ~ (12.6.2)
To correct for nonzero overlap angle, a is replaced in these equations by a + y',
where y' is the overlap angle during the fault, which is larger than the prefault
overlap angle. Thus for normal operation, where delay angles are usually less
than 30 degree, twelve-pulse operation will have smaller overcurrents than six
pulse operation.
Figure 12.9(a) also applies when the abscissa angle depicts the time between
fault inception and the previous valve firing. The twelve pulse curve wfil,
therefore, only be applicable from 0 to 30 degree, and thus twelve pulse operation
always has a lower peak overcurrent than does six-pulse operation. The negative
slope of the curves is indicative of the time which must elapse before protective
 Angle degrees
(a)

2 .o

Fig. 12.9. (a) Effect of instant of fault (and a,) on Id:; (b) Effect of
detection delay on Id:

retard action of the rectifier firing angle can be initiated. Again, instantaneous
The cumulative
detection is assumed, this time with a constant firing angle of amin.
effects of initial firing angle and elapsed time before protective action becomes
effective can be obtained from these curves.
The effect of noninstantaneous detection on the level of overcurrent is shown in
Fig. 12.9(b) for a zero resistance earth fault on the line side of the rectifier end
terminal. The nominal firing angle was 25 degrees with the fault applied 10
degrees after the last firing. Calculations based on the above idealized formulae
(12.6.1) and (12.6.2) indicate levels much higher than those obtained by
the dynamic simulation, for both instantaneous detection and for detection 30
degrees after fault application.
This indicates the need for the more accurate simulation in which the a.c.
system, filters, d.c. system parameters and converter controls are accurately
modelled, in order to assess the effect of voltage regulation, commutation, and
waveform distortion. Unless the simulation can accurately represent the actual
timing ofvalve firings in relation to fault occurrence,and the subsequent effects of
rapid control action associated with normal or protective action, inaccurate
results will be obtained.
The interactive use of the state variable dynamic analysis and a transient
stability program described in Chapter 11 provides accurate simulation of the
a.c.-d.c. system behaviour over the entire fault and recovery period.
The dynamic analysis requires suitable representation of practical d.c. line fault
detection and recovery schemes. Similarly, accurate representation of the
expected fault behaviour, including the arc characteristics, must be incorporated
into the model.

12.6. DETECTION OF D.C. LINE FAULTS

Based on the characteristics discussed in Section 12.5 a d.c. line fault detection
and protection scheme must:
(i) respond rapidly to d.c. line faults independent of converter or plant
parameters, and of fault location;
(ii) not respond to converter faults, a.c. system(s)faults, or reduced d.c. voltage
due to operational changes (e.g. step changes in d.c. voltage);
(iii) determine if converter restart has been successful.
Since, as described in Section 12.5, the protection must be fast acting, the
detection scheme must be able to identify the travelling waves set up on the
transmission lines by each type of fault, and to discriminate between them.
The polarity of these travelling wave gradients, with respect to line voltage,
holds enough information to determine whether the fault is bipolar or
monopolar, and to indicate the pole(s)involved. In the case of a monopolar fault,
travelling waves are also induced on the healthy pole by virtue ofmutual coupling
action. Investigations into possible overvoltage thereon, by Hingorad4) and
Kimbark,") have shown that in certain circumstances the coupling action as a
result of a monopolar fault may be sufficient to cause a flashover on the healthy
pole. The magnitudes of overvoltages induced on the line by coupling action are
dramatically influenced by the d.c. line terminations. Fault location and type, as
well as line characteristics, will also affect the overvoltage magnitude at the
terminals. As indicated in Section 12.5,converter control will have no effect on the
first-wave reflections at the terminals, which in general involve the peak
overvoltages.
Because of the d.c. link smoothing reactors, inverter faults and a.c. system
changes will cause considerably lower gradients of line voltage and current at the
rectifier than will d.c. line faults, since there is much more inductance between the
detection unit and the fault location. Similar discrimination is possible with
current gradients, since the two types of travelling waves initiated by the fault are
interrelated.
The fault resistance will also affect the travelling wave gradients. In the case of a
high-resistance ground fault close to the inverter end of a long line, information
obtained at the rectifier end, based exclusively on voltage magnitude and
gradients, may not be sufficiently reliable, since it may be indistinguishable from
voltage information resulting from operational changes or a.c. faults. Resorting
to comparison of current at either end is also unreliable, and would require signal
transmission. The use of current gradient, however, is a reliable signal. The
computer model uses the weighted sum of the direct voltage and direct current
gradients at the converters, for each pole of the d.c. line,',) i.e.

which will be directly related to the travelling waves initiated by the fault, and
contains information relating to the fault type, size and location. Since the
terminations of the d.c. line will affect the gradients of the travelling waves, this
must be considered in setting suitable detection levels.
By suitable choice of K, and K,, restraints may be put on q to ensure that a
monopolar fault is not seen as a fault on the healthy pole of a bipole system, due to
the voltage which may be induced thereon by mutual coupling action. Of course,
if this induced voltage is sufficient to initiate a flashover on the initially unfaulted
pole, an extremely undesirable condition, then q must be such as to initiate the
detection/protection scheme associated with that pole as well. On detection of a
line fault, the converter bridges associated with the faulted pole(s) will then
undergo the relevant protective action as described below.

12.7. FAULT CHARACTERISTICS

Arc extinction and deionization

Converter blocking is normally used only for internal station faults, and is not
necessary for d.c. line faults, where a method of fast retardation of firing angles is
most effective.
On detection of a line fault, by driving the rectifier into inversion (say a, = 120
to 135 degrees), the line voltage will rapidly drop appreciably below the line side
value, forcing the rectifier current to extinguish more quickly than it would under
normal C.C. control action. Similarly, the drop in line voltage at the inverter due
to the line fault will extinguish the inverter current quickly.
However, this condition of zero energy transfer into the link will not
immediately extinguish the fault. In the case of a d.c. link such as in the New
Zealand scheme, the combination of a long overhead line (with smoothing
reactors) and a relatively long submarine cable represents a lightly damped
oscillatory RLC circuit with a natural frequency of approximately 50 Hz. This
circuit has a significant energy storage capacity. With no external sources, i.e.
once both converters have shut down, the time for the energy in this circuit to
dissipate is determined by the relative magnitude of the circuit resistances, which
will be made up from three components:
(i) d.c. line resistance;
(ii) earth fault resistance; and
(iii) arc resistance.
The value of line resistance will be known for a particular line and therefore to
decide upon the interval necessary for complete arc extinction, the properties of
the arc, as well as the specific earth resistance, must be established. Tests carried
out by Kohler") determined the following arc voltage and the arc path
deionization times of a free burning arc:
(i) The arc voltage drop for arc lengths of 0.4 to 2.0 metres seemed to be
independent of fault (rectifier) current and of arc length, and had an
average value of 2 kV per metre of arc.
(ii) The deionization time, measured from the instant when the arc current
goes below 1A, must be from 20-50 ms to prevent reignition of the arc on
converter restart. Higher line recharge voltages were found to require
slightly longer deionization periods.
(iii) A minimum interruption time of 100-150 ms may be sufficient to restore
full d.c. transmission if the fault is of a temporary nature.
In these tests by Kohler, the d.c. line was not simulated, so that the only storage
element involved was the d.c. smoothing reactor. For this reason, relatively quick
extinction was achieved-of the order 20-30 ms. However, when the line storage
elements are represented, the extinction time will be longer, depending on the
decay of the energy stored in it.
The values used for earth resistance will then greatly affect the time necessary
to achieve this current extinction, since as discussed above, the magnitude of
resistance in the fault circuit will determine how quickly the naturally oscillating
fault voltage (and fault current) envelope decays. In general, earth resistances of at
least the magnitude of the tower footing resistance must be included, except that
the towers may be connected in parallel by earth wire. Typically values between 5
and 20 i2 can be expected.
Since, as detailed above, the arc voltage drop is relatively constant, the arc
resistance is dependent on the instantaneous arc current. Therefore, the arc
resistance must be calculated as a function of the instantaneous arc current.
The fault impedance is predominantly resistive and therefore the fault voltage
and current will be in phase. With a naturally oscillating current waveform, the
period ofwhich is dependent on the line parameters (normally in the range 10-60
Hz), there will be several instances at which the arc may extinguish. However, if
the voltage impressed across the still ionized arc path has sufficient peak
amplitude to sustain further arc current, i.e. the voltage at the fault location is
greater than 2 kV per metre of arc, then the arc path will continue to conduct
current. Once the maximum value of voltage (line-earth value) is insufficient to
re-ignite the still ionized arc path after current has passed through zero, then the
current will remain at zero. The deionization time, representing the period taken
for the arc path to deionize, will then be measured from this time on.
The majority of h.v.d.c. schemes, including the New Zealand scheme con-
sidered here, have higher direct voltage and direct current ratings than the test
system used by Kohler. Therefore, Uhlmann's@) more conservative recom-
mendation of 100 ms is used as a safe deionization interval in this chapter.
After this time, the arc path will generally be sufficiently deionized to withstand
normal voltage. During the arc extinction and deionization periods, the rectifier
maintains a firing angle in the range 120 to 135 degrees. Normal inverter C.C.
control action would have driven the inverter delay angle towards its minimum in
an attempt to maintain link current at K:!.
In the New Zealand scheme, the presence of a large capacitance (submarine
cable) at the end of a long overhead line can create line voltage oscillations when
the inverter firing,angle is advanced (by C.C. control) towards 90 degrees and
these oscillations can force the line voltage to change polarity following line fault
occurrence. Furthermore, resonance effect between the d.c. line and the control
system may sustain the line voltage oscillations, thus extending arc extinction
times. A protection scheme which overrides the inverter C.C. control and
maintains E.A. control reduced this problem considerably.

Converter restart
On completion of the deionization period, the program initiates the restart
procedure for the link. In most cases, the line is restored to normal voltage and
prefault power. If re-energization at full voltage is not acceptable (e.g. due to wet
or dirty insulators), then a lower voltage may be used by bypassing one or more of
the converter bridges, or by the use of constrained control action, i.e. larger
minimum limits on a, and y,.
Transmission cannot restart itself, since operation at the completion of
deionization is with both converters in inversion, i.e. a, = 125" and y, =yo. A
starting order is therefore needed to remove the protective control action and to
release the normal control systems of the converters. The actual time required for
the d.c. link to regain nominal voltage and current will depend on the properties
of the d.c. line and the converter controls. Generally, fast restart with a speed of
power recovery conducive to stable operation, but with avoidance of overswings
in both a.c. voltage and current, is required. Oscillations of these parameters may
otherwise occur at the natural frequency of the line, and may be sufficient to
cause arc restrike or to put additional stress on other plant. For this reason,
restart by an instantaneous release of the normal controllers is not practical.
To avoid overswings in d.c. voltage (and current), it is recommended to
increase the direct voltage exponentially to the nominal value, with minimum
overshoot, rather than as a sudden step. This unit usually operates only for start-
up or re-energization following a line fault. The time constant of the exponential
increase must be long compared to the natural frequency of the line (e.g. 100 ms,
cf. 20 ms), to ensure stable, controlled recovery of the d.c. system from the fault
condition is obtained. A similar method is used in practice on the New Zealand
scheme with an equivalent time constant near 100 ms.
If the line fault persists, or the arc restrikes due to inadequate deionization, the
above process may be repeated a few times with increased deionization periods
and/or reduced restart voltages, and then, if it still persists, the faulted pole
removed from service.

12.8. RESULTS

Fault application
Once satisfactory initial conditions for the test system are achieved (requiring
about two cycles of dynamic simulation) a pole-to-pole short circuit adjacent to
the line side of the rectifier smoothing reactor is applied, immediately after a valve
firing in one of the rectifier bridges. Thus, for twelve pulse operation the fault
occurs 30 degrees before any retard action becomes effective. Initiation of the
fault is implemented by alteration of the relevant connection matrix element (K,,
or K,,) for an already established branch element, or by insertion of a new branch
element (R,) and reformation of K,,.

Fault detection
In the state-variable formulation, the detection principle is implemented by
monitoring the rates of change of the state variables I//, and Q, on each pole of the
d.c. line at each converter terminal, where 1 is the smoothing reactor branch and a
is the node relating to the converter's line side termination. Equation (12.6.1) may
then be rewritten as

An indication of the relative detection times for the two converter units is
shown in Figs. 12.10(a) (converter direct currents) and (b) (line side converter
voltages). An indication of the travelling wave initiated by the fault can be
obtained from this graph, as can the natural frequency of the d.c. line. The rectifier
detection unit is seen to respond almost instantaneously to the fault, as the
voltage gradient is very high. The protective action initiated, with the next firing
pulse retarded, is seen to be effective after 30 degrees, as predicted. This is
indicated by a change in the rate of rise of rectifier current. The time delay
associated with the travelling waves on the line is indicated by the inverter
detection delay of around 5 ms between fault initiation and identification.
Inverter current is seen to extinguish after approximately 8 ms, and the rectifier
current 6 ms later. The oscillation of inverter line voltage is seen to be of about 50
 Time (cycles)
(a)

-0.5L 1 1 I I I I I
2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 i
Time (cycles
44 C
$.$ g .$E%
-s'esg'=
L0
s'=
zag$
- 0:
""'0
r
w c, (b)

Fig. 12.10. Coordination of d.c. line fault detection action-detection

responses

Hz frequency. The envelope of the fault voltage is seen to have a peak of around
0.15 p.u. or about 30 kV, which will promote a healthy arc.

Arc extinction and deionization

Prior to rectifier shut-down (which takes about 14 ms), the earth resistance has
little influence on the fault voltage and energy state of the d.c. line.
Subsequent to rectifier turn-off, the only fault current source is the energy
stored in the d.c. line, and the fault current will thus decay exponentially with a
time constant dependent on the total circuit resistance.
Studies for earth resistances of 2,5,10 and 20 ohms were conducted; a typical
fault current is illustrated in Fig. 12.11 and the extinction times achieved are
summarized in Table 12.1.
 -1.51 I I I I
2.0 3.0 4.0 5.0 6.0 7
Time (cycles)

Fig. 12.1I. Earth fault current

Table 12.1 Arc extinction times as a function of earth resistance

Earth resistance (ohms)

2 5 10 20

Arc extinction times (ms) 110 100 90 80

It is apparent that each halving of the earth resistance increases arc extinction
by approximately a half cycle. The arc extinctions occur at the zero crossings of
arc current, provided the peak arc voltage is below the arc sustain level
(approximately 2 kvlmetre). The arc path resistance then increases effectively to
infinity, since no current flows, and provided restriking does not occur, the fault
branch may be removed or disconnected. This action is invoked at the
completion of the specified deionization time, when it is assumed the arc path is
completely deionized.

D.C. line recharge and power transmission restart

To recharge a d.c. line from a, = 125" and y i = yo, a, is reduced by a exponential
control function, while maintaining y i = yo. This ensures the rectifier recharges
the line fully before the inverter begins to draw power. This will mean that both
the rectifier and inverter constant current controllers will be inoperative, since the
inverter is on E.A. control and the rectifier on the start control function.
Following the completion of the deionization period, a, is stepped from 125 to
90 degrees over one firing instant. Subsequent firing action is then controlled by
the recharge function
 Recharge function (alpha
100.0

00.0
0.0 1.0 2.O 3.0 4.0 5.0 €
Time (cycles)

Fig. 12.12. Characteristics of exponential recharge k values: (i) 2.0 ;

(ii) 2.4; (iii) 3.0; (iv) 4.0; (v) 5.0

where
a, is the control angle which will give nominal line voltage, and
At is the elapsed time since transfer of the converter control to the restart
unit (i.e. from when a, = 90").
The constant k will determine the response rate.
Figure 12.12 illustrates the variation of a, for various values of k, as a function
of elapsed time (At).In this case a, = 10". For a d.c. line which has a time constant
of approximately 20-25ms, it is desirable to have no more than 63 percent of
nominal voltage when At = 25 ms, in order to obtain critical damping of the no-
load voltage rise. Curve (ii) exhibits this characteristic, and actually achieves
nominal voltage when At is approximately 100 ms.
The recharge process is thus implemented by adjusting the firing instant F j + l
by

where

and where At is measured to F + . ,

When the d.c. line is fully recharged to nominal voltage at the correct polarity,
conditions are conductive to resumption ofnormal power transfer. At this time a,
will be close to nominal and 6, will be on its specified optimum value. The rectifier
then switches from the line recharge controller to C.C. control, whilst the inverter
continues on E.A. control.
Full simulation of a d.c. line fault
The temporary bipole fault under consideration is now followed from the instant
of fault initiation through to full fault recovery and normal power transmission,
using the TCS and TS programmes described in previous chapters according to
the protection philosophy discussed in the above sections. A flow diagram
indicating the coordination of the special control functions used for protection
and recovery of d.c. line faults is shown in Fig. 12.13. The results are indicated in
Figs. 12.1qa)-(g).

Detection

I Fault present
1

V NO
4

{ ~ i n erecharg! in pragr
YES

1
.t-
I Set a, = 125' 1
or Y l = YO
render
controllers
inoperative
1

NO
{
1 'I
Extinction

Deionization

Fault removal
i
1 Remove fault
JYES
1

1
NO

r
-
YES
a, > 90'
t
NO Set a,= 90'

D.C. line 1 1

Line rechar ed -
restore CC! regulators
to both rectifier ond
mverter

@and@ represent e x l and re-entry points from t k dynamic

simulations main controlling subroutine

Fig. 12.13. Flow chart of coordination of special control characteristicsfor d.c.

line fault protection
Time (cycles)
(a)

160 18.0 2
Time (cycles)
(b)

0
Time (cycles)
(c)
 Time (cycles)
(d)

-1.51
0.0 2.0 4.0 6.0 8.0 10.0 120 14.0 16.0 180 2
Time (cycles)
(el

25.0

12.5

0.0

-12.5

-25.0

Time (cycles)
(f)
 12.5

0.0

-12.5

- 25.0
0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0 18.0 20.0
Time (cycles)
(g)

Fig. 12.14. (a) Converter d.c.currents ;(b) Converter d.c. voltages ;(c) Converterd.c. powers ;
(d) Rectifier a.c. voltages; (e) Inverter a.c. voltages; (0 Rectifier a.c. currents;(g) Inverter a.c.
currents

The fault was applied and detected using the process described in Section 12.8.
As a, is retarded to 125", Fig. 12.14(f) shows that two phases at the rectifier end
continue to conduct current for an extended period. The initial rectifier current
peak, which occurs 4 ms after the fault is applied, is only 80 percent above the
setting and decays to zero, 15 ms after fault initiation. The inverter has already
shut down by this time. Figure 12.14(b) illustrates the existence of voltage
oscillations, with the inverter voltage reaching 50 percent of the rated voltage
with reverse polarity. The oscillations die out within approximately 80 ms. Due to
the continued presence of the harmonic filters, a residual current level remains
throughout the fault (Figs. 12.14(f)and (g)), with considerable initial harmonic
content causing corresponding a.c. voltage waveform distortion (Figs. 12.14(d)
and (e)).This is more evident at the rectifier end due to the large initial increase in
direct current and also due to the large change in control angle, which does not
occur at the inverter end under E.A. control. The voltage distortion is seen to
reduce quickly as the residual (filter) currents become settled. In each case this
takes about five cycles, giving an indication of the relatively slow time response of
the harmonic filters.
During the same period the line voltages are seen to decay, and as indicated in
Fig. 12.14(b), extinction of the arc occurs some 80 ms after fault initiation.
During the deionization period, which extends for approximately 100 ms, the
a.c. voltage waveforms become sinusoidal but some amplitude variation is
noticeable in Figs. 12.14(d) and (e) as a result of the time-variant Thevenin
sources.
Some 200 ms after fault initiation, the arc is expected to be completely
deionized, the fault branch is removed and fault recovery commences. The
rectifier direct current and both line voltages indicate the effect of the exponential
recharge function. The current involved in line recharge is seen to be relatively
small. Because the inverter is o n E.A. control, the inverter does not conduct
current until the line is charged to about 90 percent of nominal. The recharge
time, measured from deionization until the line is recharged and converter
current has begun to increase, is seen to be about 80 ms.
Following the recharge period, which is completed 280 ms after fault initiation,
the converter currents are seen to increase quickly and steadily to the set values.
This increase takes place with the rectifier on C.C. control, and the inverter on
E.A. control. The current is seen to reach its setting after 60 ms with only a minor
initial overshoot, indicating that the C.C. control constants used are adequate. At
this time (cycle 19), the d.c. power levels are not a t their prefault values since the
inverter a.c. voltage is still depressed, although it is increasing steadily with the
reinjection of link power.

12.9. REFERENCES
1. J. Arrillaga, M. D. Heffernan, C. B. Lake and C. P. Arnold, 1980. 'Fault studies in a.c.
systems interconnected by h.v.d.c. links', Proc. IEE, 127, Pt. C (l), 15-19.
2. H. A. Peterson, A. G. Phadke and D. K. Reitan, 1969. 'Transients in e.h.v.d.c. power
systems: Part I-rectifier fault currents', Trans. IEEE, PAS-88 (7), 98 1-989.
3. M. D. Heffernan, J. Arrillaga, K. S. Turner and C. P. Arnold, 1980. 'Recovery from
temporary h.v.d.c. link faults', Pans. IEEE, Paper 80SM 67.5-679. PES Summer
Meeting, Minneapolis.
4. N. G. Hingorani, 1970. 'Transient overvoltages on a bipolar h.v.d.c. overhead line
caused by d.c. line faults', Trans. IEEE, PAS-89 (4), 592-610.
5. E. W. Kimbark, 1970. 'Transient overvoltages caused by a monopolar ground fault on
bipolar d.c. line: theory and simulation', Trans. IEEE, PAS-89 (4), 584-592.
6. A. Erinmez, 1971. 'Direct digital control of h.v.d.c. links.' Ph.D. thesis, University of
Manchester.
7. A. Kohler, 1967. 'Earth fault clearing on an h.v.d.c. transmission line with special
consideration of the properties of the d.c. arc in free air', Trans. IEEE, PAS-86 (3), 298-
304.
8. E. Uhlmann, 1960. 'Clearing of earth faults on h.v.d.c. overhead lines', Direct Current, 5
(2), 45-57, 65-66.
 Transient Stability Analysis of
A.C.-D.C. Systems
13.1. INTRODUCTION
Having developed a detailed transient converter simulation program (TCS)
suitable for analysing the behaviour ofan h.v.d.c. link just after a disturbance, it is
only necessary to merge this model with a transient stability (TS) program to
complete the suite of programs. It is then possible to examine the response of all
system components over a period of several seconds.
Many of the problems associated with the merging of the two programs have
been discussed in Chapter 11. This was done to ensure that acceptable a.c. system
information was made available to the TCS programme. This chapter firstly deals
with the processing of the information, as obtained from the TCS, into a form
suitable for inclusion in the TS programme.
The second part of this chapter is concerned with three transient stability
analysis studies of the New Zealand a.c.-d.c. power systems. These studies are
used to illustrate principles as well as to justify the use of the integrated programs.

13.2. REQUIREMENTS O F THE TRANSIENT STABILITY PROGRAM

Transient stability (TS) studies are normally carried out on the assumptions of
balanced and sinusoidal waveforms and in the presence of h.v.d.c. links such
assumptions need to be extended to the voltage and current waveforms at the
converter terminals. As explained in Chapters 11 and 12, part of the study
requires detailed transient converter simulation, the results of which cannot be
directly incorporated in a TS study.
It is therefore necessary to derive fundamental frequency quantities from TCS
voltage and current waveforms before information can be transferred from the
TCS to the TS study.
Two variables are needed to transfer the converter information from the TCS
to the TS study. A number of alternatives are available, i.e. real power, reactive
power, voltage magnitude and current magnitude. Various definitions have been
offered for reactive power in the presence of distorted waveform^,")(^)(^) but a
meaningful value of reactive power can only be obtained from sinusoidal
components of voltage and current of the fundamental frequency. This funda-
mental frequency reactive power can only be derived from spectral analysis of the
voltage and current waveforms and since both these variables can be used on
their own, reactive power becomes redundant. The above problems do not exist
in the definition of real power and further justification for its use as a variable is
given later.
The choice of the second variable is between voltage and current magnitude.
Since the converter acts as a current harmonic source. the current waveform,
particularly during and immediately after a disturbance, exhibits greater
distortion and modulation effects than the voltage waveform. This is illustrated in
the typical waveforms of Fig. 13.1.

Time (cycles)
(a)

-20 0
:1 5 13.5 145 15.5 165 115 185 19.5 205 215
I
22.5
Time (cycles)

(b)

Fig. 13.1. Comparison between voltage and current waveforms at the rectifier terminal
during current setting changes (from a TCS study). (a) Rectifier voltages; (b) Rectifier
currents
 The spectral analysis of the voltage waveform is, therefore, less prone to error
and the voltage magnitude is a better choice of variable.
A specification of voltage and real power is required at each integration step of
the TS study when the quasi-steady-state d.c. link model is replaced by the results
of TCS.
To obtain fundamental components of voltage and power, the TCS waveforms
must be subjected to Fourier processing over a discrete number of fundamental
cycles of the waveforms. A component of voltage and power can be obtained at
any sample point using a complete cycle (an aperture size of period T) spanning
T/2 both sides of the sample point as illustrated in Fig. 13.2(a). The sampling
points must be synchronized with the TS study integration steps and, in order to
ensure that sufficient samples are usedJ4)the TS program step length should not
exceed half of one fundamental cycle. It is convenient, therefore, to fix the step
length and sample rate at half a cycle of the fundamental frequency as illustrated
in Fig. 13.2(b).
Cycle of ~ n t e r e s t ~
t it--
: , TCS waveform
\

I sokple point for T S

(a)

Seauence of

I
I
I
I

Aperture for
somple points
(b)

Fig. 13.2. Interaction between TS sample points and a

TCS waveform. (a) TCS waveform; (b) Sample points

13.3. OBTAlNING INFORMATION FROM THE TCS PROGRAM

The voltage waveforms at the terminals of a d.c. link during typical disturbances,
such as those illustrated in Fig. 13.3 cannot be interpreted in terms of simple
periodic waveforms.
The reasons for this are:
(i) Converter control action introduces deviations in the equidistant firing
regime and valve firings do not necessarily occur periodically.
 T ~ m e(cycles
( 0 )

-1 21
00 06 12 18 24 30 36 42 48 54
Time (cycles)

(b)

Fig. 13.3. Typical waveforms obtained from a TCS study. (a) Inverter voltage
waveform for a three-phase fault at the inverter Terminal; (b) Rectifier voltage
waveforms for a single-phase fault in the inverter a.c. system

(ii) The nonlinear behaviour of the converter excites disturbances which are
not periodic in T.
(iii) The network equivalent source is given in a constant frequency frame of
reference but the TS study deviates from constant frequency after a
disturbance, effectively altering the fundamental period.

Spectral analysis
The above effects can be classified in terms of modulation, frequency mismatch
and nonperiodic noise. Their presence means that a simple fundamental
component of voltage (or current) cannot be accurately obtained merely by using
a standard discrete Fast Fourier Transform (FFT),(5)on the samples which make
up the TCS waveforms of Fig. 13.3.
 -1.5 u
2.0 2.2 2 2.6 2.8 3.0
Time (cycles) Time (cycles)
(a) (b)

Fig. 13.4. Single cycles of one-phase obtained from TCS waveforms of Fig. 13.3 (a)
Inverter voltage; (b) Rectifier voltage

Figure 13.4 illustrates single cycles of distorted waveforms. The discontinuities

at the boundaries of these individual cycles show that considerable levels of
nonperiodic frequencies exist in the waveforms due to the nonlinearities
associated with control action, filter response and converter behaviour. If these
waveforms are assumed to be periodic and are subjected to Fourier transfor-
mation to obtain the spectral components of the signals (in particular the
fundamental component), the phenomenon known as spectral leakage(6)occurs.
Spectral leakage results in the nonperiodic noise contributing to each of the
periodic spectral components present, introducing uncertainty in their identifi-
cation by the Fourier transformation.
Work(7)has been done to determine the magnitude of the contribution of the
non periodic noise. The results of over 100 different cycles of TCS waveforms
indicate differences of up to 0.7 percent for very distorted waveforms. The
maximum difference observed is 1 percent for the cycle illustrated in Fig. 13.4(b).
The differences are small and it can therefore be concluded that spectral
leakage does not contribute significant errors when identifying the fundamental
components of TCS waveforms.
It is therefore possible to obtain fundamental powers and voltages from TCS
waveforms by simple application of an FFT, in the presence of modulation,
frequency mismatch and non periodic noise. This can be done with the same
degree of accuracy expected from other input data in transient stability studies.

Simplification in obtaining results

An appro~imation(~' can be used to reduce the effort of performing spectral
analysis of the TCS waveform.
Apparent power is defined as:
The real component of s(t) is the average value of s(t) taken over a period, T, i.e.

This is readily evaluated from discrete signals using:

where N is the number of samples.

The real power evaluated in this way is the sum of fundamental and harmonic
real powers and the results(*)from many TCS's show that, in general, the har-
monic real power is small and that the rms approximation for real fundamental
power is a good one. Figure 13.5 shows two representative curves. In the case of
Fig. 13.5(a), although the maximum error is 4 percent, thecummulative error over
the simulation period is small. The maximum error occurs for a short period
when highly distorted signals exist.
The harmonic power flow can only result from in-phase components of

-8.0 I I I I I I I I
0.0 1.2 2.4 3.6 4.8 6.0 7.2 8.4 9.6 10.8 1:
Time (cycles)

-2.0 -

0.0 0.6 1.2 1.8 2.4 3.0 3.6 4.2 4.8 5.4 r
Time (cycles)
(b)
Fig. 13.5. Comparison of RMS and fundamental powers for the cases
shown in Fig. 13.3. a) Three-phase fault at the inverter end b) Single-
phase fault at the inverter end
harmonic currents and voltages. Considering the relatively low resistive com-
ponent of the converter transformer and the large X I R ratio of the a.c. system, the
in-phase component of the harmonic currents and voltages are very small.
Therefore the use of rms power is a good approximation and spectral analysis is
not needed to derive the power variable.
The rms value of a voltage, V(t), is defined over a period T as

For discrete data this becomes:

where N is the number of samples.

Figure 13.6 shows a comparison of fundamental voltage (positive sequence)
and rms voltage for the TCS waveforms of Fig. 13.3. Figure 13.6(b) represents a

Time (cycles)
(a

Time (cycles)
(b)
Fig. 13.6. Comparison of RMS and fundamental voltages for the TCS
waveforms of Fig. 13.3. (a) Inverter voltage; (b) Rectifier voltage
severely distorted case with a 5 percent difference observed during the period
immediately after disturbance initiation. This coincides with the period of
maximum distortion in Figure 13.3(b) at which time harmonic content is high.
This error is too large to be considered insignificant when transferring data from
the TCS to the TS programme.
It is therefore necessary to retain the spectral analysis to determine a
fundamental voltage magnitude from the transient converter simulation wa-
veforms but the rms approximation for power avoids the need for spectral
analysis of current.

13.4. INCORPORATING TRANSIENT CONVERTER SIMULATION IN

A TRANSIENT STABILITY STUDY
When a detailed transient converter simulation has been performed on a system,
the TS compatible equivalents derived from the results can be used to replace the
QSS model during the most disturbed part of the study. When the response of the
d.c. link has settled the TCS is terminated. The TS study then reverts back to
using the QSS d.c. link model for the remainder of the study.

Solution at the converter node using TCS input '

The information from the TCS is embodied in real power (P,,,), and voltage,
(Ed,& variables obtained from processing of the TCS waveforms. These become
the specifications at the link terminal node when the QSS model is removed.
Using the unified algorithm developed in Section 10.5 the system equivalent
shown in Fig. 13.7 is solved, given the P,,, and Ed,, specifications.

Fig. 13.7. Equivalent system for including TCS data

This system can be used to solve directly for the a.c. currents which can then be
superimposed on the a.c. network in the same way as the a.c. current from the
QSS model.
The complex power Sf& is given by
Taking real parts of equation ( 13.4.2)and rearranging, 8 can be obtained directly
by:
~=~-P+COS-
The a.c. current can now be obtained using

This solution requires little computational effort when compared to the QSS
link model. With very low converter terminal voltages, the first term of equation
(13.4.3) becomes very large and any mismatch between Ed,, and P d y ncan cause
the cosine function to exceed unity. This difficulty is overcome by releasing the
power specification and allowing it to change until the equation can be solved.
The small adjustment of power setting has little effect on the transient stability
analysis, since the converter power involved is small under very low voltage
operating conditions.

Alignment of TCS input with TS

The TCS results may be reduced to a sequence of data points, spaced half a cycle
apart and each representing the fundamental component over the cycle spaced
about that point. The integration step of the TS study should then be fixed at half
a cycle to coincide with the data points from the TCS.
In the TS study, at a disturbance initiation, a new solution is obtained for the
system immediately after the disturbance application. However, because the TCS
results are essentially waveform orientated, a correct data point cannot be
obtained at the new solution after disturbance application. Figure 13.3(a)shows
sudden discontinuities at approximately two and seven cycles and Figure 13.qa)
demonstrates the averaging effect of processing the waveform results.
It is therefore necessary to extrapolate the processed TCS data points to
provide, as closely as possible, a representation of the link performance
immediately before or after a disturbance application. This effect is demonstrated
in Fig. 13.8 in which the discrete integration steps of the TS study are related to
the data points obtained from TCS.

Fault on Fault clearance

Fig. 13.8. Data alignment between TCS and TS program

 Fault clearance requires special consideration as the mechanism differs from
the two models. In the TS model the inherent assumption is made that at the time
of fault clearance, Tfc,all three phases clear simultaneously and a new solution is
obtained immediately (i.e. at T:). For the TCS, although the decision is made to
clear the fault at Tfc,only one phase will open within 60 degrees of Tfc(i.e. at the
first current zero), the remaining phases opening in the following 120". This
would appear to distort the alignment of the two models for the TS solution at
T:.
However, the disturbance caused by the first phase clearance has the greatest
influence on d.c. link control, and in spite of the TCS clearing delay, the post fault
TCS data, obtained by processing one full cycle after Tfc, can be used to
superimpose the immediate effect of the fault clearance on the TS solution at T:.
At the first and last points of the TCS data sequence, the QSS model is solved as
well, to provide a check on the accuracy of matching between the two programs
and the two link models.

13.5. MODIFICATION OF TRANSIENT STABILITY

PROGRAM FOR A.C. FAULTS
The modelling of a.c. faults takes different forms in the TS and TCS programmes.
For the TS program described in Chapter 9, the fault admittance is included
directly in the admittance matrix of thenetwork when the fault is to be present, i.e.
the fault bus is included implicitly in the equation

Alternatively, the fault can be separated from the admittance matrix to obtain a
solution for the fault current using the Thevenin equivalent at the faulted node.
The fault current can then be superimposed onto the network in the same way as
the injected current obtained at the d.c. link nodes.
It is necessary to solve both the fault and the d.c. link simultaneously because of
their mutual coupling'shown in equations (12.2.2) to (12.2.4).
The simultaneous solution can be achieved by reformulating the d.c. link

Fig. 13.9. Fault bus representation

model completely so that the fault equations are directly included in the Newton-
Raphson algorithm. However, this special formulation would only be required to
provide the time variant equivalent for TCS during the fault period. Thus the
resulting extra computational demand should be avoided if possible.

Algorithm for simultaneous fault and link solution

The two different models can be retained by using a simple algorithm in the TS
program which involves the simultaneous solution of the d.c. link model and the
fault during the period when it is applied.
The solution of the fault is obtained by using the simple network of Fig. 13.9

Solution of machines and

Network voltage without

fault 0rd.c. link included equivalent for TCS

.1
due to d.c. link obtained

Estimateof Theveninvdtoges

New estimate of Thevenin

Solve d.c. link and

superimpose on [v,,,,]

combined with d.c. linkand

Fig. 13.10. Flow chart for simultaneous solution of

fault and d.c. link
where V pand Z p pare the Thevenin voltage and impedance at the fault bus. The
fault current is obtained from:

For this to be a correct solution, the Thevenin theorem states that all other
voltage and current sources must be present, i.e., the network equivalent at the
fault bus must be obtained with the d.c. link injected currents included. The
converse also applies in obtaining the correct solution for the d.c. link. The
solutions of both link and fault are combined in the algorithm of Fig. 13.10.
The algorithm relies on a good estimate of the d.c. injected currents at the first
iteration of each step if the speed of convergence of the TS program is not to be
impeded. However, the link solution obtained at the previous step is suitable
except for the first iteration after fault application. In this case, since the fault has
a dominant effect on the network, it is solved alone at the first iteration. A new d.c.
link solution can then be obtained for entry into the algorithm (at A) of Fig. 13.10.
This starting process is illustrated in Fig. 13.11.
The provision of a time variant system from the TS program is much simpler
with a d.c. fault. For this case, it is not necessary to provide explicit fault bus
representation since the fault is contained within the d.c. system being simulated.
The equivalent provided for the TCS program in this case is exactly the same as
that obtained for the QSS d.c. link model.

First solution of machines and network

t
Solve for fault current 1
using [v,] [ 11
-

Solve d.c. link and

Fig. 13.1 1. Starting process after

discontinuity
13.6. RESULTS USING THE COORDINATED PROGRAM
In this section, the a.c. and d.c. system responses to typical disturbances, using the
coordinated program, are compared with those obtained using the QSS
converter model. From the results, it is possible to assess the validity of the QSS
model used in the TS program. Techniques adopted to ensure accuracy and
economy during the running of the coordinated program are also discussed
wherever necessary.
The New Zealand a.c. power system used for these studies consists ofthe South
Island primary system as shown in Fig. AII.2 and a simple Thevenin equivalent
for the North Island primary system in parallel with the synchronous condensers
at the inverter terminals. The QSS model represents each d.c. link converter by a
single bridge equivalent while for the TCS the d.c. link is represented using a
twelve-pulse configuration as illustrated in Fig. 13.12.

South Island a c system North Island

~ d t i h i ; e ' n ~ n Fault bus Smooth~ng Cable Over- Smoothing a
, reactor head reactqr
; ;SIX-element 11ne j I
I I overhead lme

Filte

Fig. 13.12. D.c. link representation for transient converter simulation

In this representation, the transient stability program, using a QSS d.c. link
model, provides the time variant system equivalents for the TCS. In turn the
instantaneous values of I,, I , , E l and E , of Fig. 13.12 are obtained from the TCS
to provide r.m.s. power and fundamental frequency voltage at the converter
terminal bus, for use in the transient stability programme.
Three worst case studies consisting of a rectifier a.c. system (South Island) fault,
an inverter a.c. system (North Island) fault and a d.c. system fault are e~aluated!~)

Rectifier a.c. system fault

A three-phase fault of five cycle duration is applied at the Twizel bus of the South
Island a.c. system (see Fig. AI1.2). The fault produces a 25 percent drop in a.c.
voltage at the rectifier terminal immediately after fault application but represents
a severe disturbance to the South Island system because of its position.
The voltage and reactive power profiles during the study are given in Fig. 13.13.
 r&d
I /
1.0-
0.9-
0 8- (3)
0 . 7-
06-
----.a,-( 1 )
&
0 5- ( 1 ) QSS
0.4- ( 2 ) TCS f~xedsystem equivalent
( 3 ) TCS time vortont system equivalent
0.3-
0 2- Fault on Fault off
0.1- t t I I I +
0.1 0.2 0.3 0.4 0 . 5 T~me(seconds)

( seconds )

h v ( 1 1QSS
-5 0 Foult off ( 2 ) TCS fixed system equ~valent
( 3 ) TCS tlme var~ontsystem equ~volent

Fig. 13.13. Voltage and reactive power at the rectifier for the
rectifier a.c. system fault. (a) Terminal ax. voltage;
(b) Reactive power

Three sets of results are shown, the first utilizes the QSS model alone without any
use of TCS. The second uses the TCS together with the TS simulation but there is
no coordination between the two parts of the program and thus fixed a.c. system
equivalents are used in the TCS. The third set of results is obtained when the two
simulations are coordinated and the TCS has a time variant a.c. system
equivalent.
For the second case, the power factor of the injected current is forced to change
from lagging to leading as a result of specifying voltage and real power at the
converter node. This results in the impossible condition of the converter
supplying reactive power to the system. The unsuitability of this form of
modelling is further emphasized during the postfault period where the TCS
recovery voltage considerably exceeds that of the dynamically responsive system.
The third set of results using interactive coordination gives a much better
match between the QSS model performance and the TCS results. The departure
between the two models immediately after fault inception and fault removal is
due mainly to the different filter models. In the QSS model, filters are represented
by a shunt admittance while the TCS reflects the full dynamic response of the
filters. For the TCS representation, the combination of the dynamic filter
response and control action results in an initial drop in reactive power demand
from the a.c. system. The mismatch is more pronounced following fault clearance.
A further cause for mismatch is the requirement to align the TCS results with
the TS integration steps, at the disturbance times, as shown in Fig. 13.8. This
distorts the behaviour of the TCS results immediately after fault removal.
Although the reactive power profile is useful in indicating the accuracy of
matching between the two programes, it is the real power transferred by the d.c.
link which affects transient stability analysis. Fig. 13.14 shows the real power flow
at the link terminals. The inverter power (Fig. 13.1qb)) is shown with positive
power flow out of the link terminal to facilitate a comparison between the two
ends of the link.

Start TCS ( 2 ) End TCS

I

5.0.
t
4 5-
4.0 -

-
a
3.5-
- ( 1 ) QSS
3 3.0 ( 2 ) TCS fixed system equivalent
( 3 ) TCS time variant system
~'25- equivalent

Fault on Fault off

0.5 1

4 Start TCS Fnd TCS

eauivalenl

Fault on Fault off

0.5
1
I I I I I -
0.1 0.2 0.3 0.4 0.5-
Time (seconds)

Fig. 13.14..Real power flows at d.c. link terminals for the

rectifier a.c. system fault. (a) Rectifier; (b) Inverter
 Rectifier
I I
0.0- I I

--
I I
I

-5.0
Inverter I

Time (cycles

Fig. 13.15. Converter d.c. currents for the rectifier a.c. system fault

In the postfault period in particular, the fixed equivalent simulation (curve 2),
departs considerably from the time variant case (curve 3) and may be omitted
from further discussion.
Immediately after fault inception, the TCS indicates that the rectifier d.c. power
flow is less than the level predicted by the QSS model. This is caused by the
sudden drop in the rectifier d.c. voltage, giving rise to a drop in d.c. current as
illustrated in Fig. 13.15. Control response at the inverter restores the d.c. current
and the link recovers from the initial disturbance quickly.
At the inverter end the immediate power response predicted by the QSS model
deviates considerably from the TCS solution, which reflects the delay effect
caused by the d.c. line.
Although instantaneous power differences are observed, the nett energy
transfer as calculated by the TCS and QSS models are similar, thus causing
relatively small first swing maximum angle differences as shown in Table 13.1.
For the high-inertia machines at the rectifier terminals (Benmore) the differenceis
3 percent, while for the low-inertia synchronous condensers at the inverter
(Haywards), the difference is 12 percent. At both terminals the QSS model gives
the larger maximum angles, and thus provides pessimistic but safe results.
The solutions of the TCS and QSS models are not exactly coincident at the
beginning and end of the TCS study as shown in Table 13.2. The results are

Table 13.1. Maximum swing angles of synchronous machines at the d.c. link terminals for the
rectifier a.c. system fault

Maximum swing angle

Benmore Haywards
Case (rectifier) (inverter)

(1) QSS
(2) TCS fixed equivalent
(3) TCS variant equivalent
 Table 13.2. Coincidence of solutions at start and end of TCS for the rectifier a.c. system fault

Link power (p.u.) Link a.c. voltage (p.u.)

Rectifier Inverter Rectifier Inverter

Qss 5.00 - 4.76 1.025 0.980

Start
TCS 4.99 - 4.78 1.022 0.979
QSS 4.63 - 4.42 0.964 0.965
End
TCS 4.51 - 4.28 0.955 0.963

obtained at the last integration step of the TS program for which a TCS data
point is available. A QSS solution of the link is obtained at this step but the
injected a.c. currents are derived from the final TCS data point (refer to Fig. 13.8).
In this case, the mismatches are small and maximum mismatch occurs for the
inverter power and is approximately 3 percent of the QSS value.

Inverter a.c. system fault

Faults at inverter terminals provide a severe test for the d.c. link controls and
often result in repeated commutation failures causing considerable disruption to
the normal valve firing sequence at the inverter. This form of converter
malfunction is of particular importance because it cannot be represented by the
QSS model and a transient converter simulation is essential.
To provide a worst case study, a three-phase fault of five cycles duration is
applied very close to the inverter terminals. Referring to Fig. 13.12, the fault is
applied to Bus 2 and a small fault resistance (1 ohm) is included to provide
sufficient source voltage for commutation to be maintained. The fault causes an
80 percent drop in voltage at Bus 2 immediately after fault inception.
Since the fault is applied at Bus 2, and the transient converter simulation
models it explicitly, the TCS currents (I, of Fig. 13.12), include the fault current
and the fault need not be applied in the a.c. system. Because the fault is resistive,
-

-8.0I I I I I I I I I I
0.0 1.2 2.4 3.6 4.8 6.0 7.2 8.4 9.6 10.8 12.0
Time ( cycles
Fig. 13.16. A.c. and d.c. apparent power at inverter terminal for the
inverter a.c. system fault
 0.0 1.2 2.4 3.6 4.8 6.0 7.2 8.4 9.6 10.8 12.0
Time (cycles 1

Fig. 13.17. A.c. and d.c. apparent power at rectifier terminal for the
inverter a.c. system fault

1 1 I I I I ,
0.1 0.2 0.3 0.4 0.5
Time (seconds)
(a
A
4.0 - Start TCS End TCS
3.0- ( 1 QSS

--z
U
2.0-
4I!
( 2 ) TCS

0- 1.0-
ii
0 I I I ,
0.3 0.4 0.5
-1.0' Time (seconds)

Fig. 13.18. Voltage and reactive power at the inverter for the inverter a.c.
system fault. (a) Terminal voltage; (b) Reactive power flow
the real power obtained in deriving the TCS equivalents, includes .the real power
flowing into the fault. The difference between the a.c. power flowing into Bus 2,
and the d.c. power flow at the inverter is illustrated in Fig. 13.16.
Immediately after the fault is applied, the inverter experiences repeated
commutation failures which persist during the fault period. The commutation
failures produce oscillations in the d.c. power at the inverter terminal, as shown in
Fig. 13.16, illustrating the effect which they have on the d.c. link performance. A
comparison between Fig. 13.16 and 13.17 shows that thed.c. power at the rectifier
terminal is smoothed by the d.c. line response. The oscillations in Fig. 13.16 are
therefore not due to the line.
Figure 13.18 illustrates the difference between the QSS and TCS a.c. voltage
and reactive power at the inverter terminals. This case has a similar filter response
to that for the rectifier a.c. fault study. In the post fault period the reactive power
demands of the link are lower in the TCS results because of the slower link
recovery for this model.
A good match is obtained for the voltages, during the fault period becaclse of
the dominance of the fault in the TCS results. In the postfault period the match is
not so accurate because of the slower link response determined by transient
converter simulation.
The active power flows at the two terminals of the link are shown in Fig. 13.19.
The QSS model shows significant departures from the results obtained using
TCS. The differences between the results are caused by three factors:
(a) Due to the d.c. line response, power continues to be fed into the rectifier end
of the line for one cycle after the fault occurs at the inverter bus.

I .
i!
-3.0
- li ( b )
il
Fig. 13.19. Real power flows at the d.c. link terminals for
the inverter a.c. system fault. (a) Inverter; (b) Rectifier
 At the inverter terminal, immediately before fault clearance takes place (i.e.
the first detected current zero after breaker operation), a commutation
failure occurs. This converter disturbance delays the clearance time of the
fault. It is impossible for a QSS model to detect this sequence of events.
In addition to the delay due to the commutation failure, inverter recovery is
also affected by the response of the d.c. line and the controller at the rectifier
end. As shown in Fig. 13.19(b)the TCS rectifier response is slower than the
QSS response.
The accurate response obtained from the transient converter simulation affects
the results of the TS analysis to a greater degree than for the rectifier a.c. system
fault study, as shown by the swing curves in Fig. 13.20. In this case, the maximum
swing angle obtained using the QSS model is less than that obtained using the
TCS model (i.e. the results are optimistic) and this is especially noticeable for the
low-inertia machines at the inverter terminal.

-901 I I I I I
0.1 0.2 0.3 0.4 0 . 5 ~
Time (seconds)

Fig. 13.20. Swing curves of Haywards (inverter) synchronous machines

for the inverter a.c. system fault

The QSS model would have to be modified to match the response

rate determined from the TCS before it could produce more accurate results.
However, accurate matching of the QSS and TCS responses cannot be generally
applied to this disturbance because of the indeterminate nature of consequential
converter disturbances and their effect on normal valve firing sequences. In
addition, the QSS model cannot predict the oscillatory response at the rectifier
terminal, as illustrated in Fig. 13.19(b) during the fault period. This is due to the
rapid rise in d.c. current in response to the collapse of d.c. voltage at the rectifier
terminal. The oscillation, however, does not involve any significant nett energy
transfer and thus does not affect the swing angle at the Benmore (Rectifier)
generator significantly.
The peak swing angle at the Haywards (Inverter) synchronous machines
occurs at 0.29 sec. and the voltage waveforms are close to sinusoidal at this time.
This is thus a suitable time to terminate the TCS. The maximum difference
between the TCS a n d Q s s results at termination is 7 percent for the inverter
power. Table 13.3 records the differences between the TCS and QSS variables.

Table 13.3. Coincidence of solutions at start and end of TCS for the inverter a.c. system fault

Link power (p.u.) Link a.c. voltage (p.u.)

Rectifier Inverter Rectifier Inverter

QSS 5.00 - 4.76 1.025 0.980

Start
TCS 5.02 - 4.79 1.024 0.979

Qss 4.27 - 4.04 0.981 0.840

End
TCS 4.12 - 3.76 0.990 0.838
--

D.C. line fault

This study is of a short-circuit fault applied on the d.c. line just beyond the rectifier
smoothing reactor and initiated immediately after a valve firing. There is thus no
power transfer during the fault period. It is a worst case study since control action
influence is delayed for 30" until the next firing.
The QSS d.c. link model has severe limitations in its ability to represent this
type of disturbance. A d.c. fault, using this model, is represented by instantaneous
shutdown of the livk at fault initiation and restart when the fault is removed.
Moreover, it yields unrealistic current peaks as it cannot accurately represent the
timing of valve firings in relation to the fault occurrence, or the subsequent effects
of rapid control action. It is also inadequate in representing the transient
imbalance of harmonic currents between the converters and the d.c. system
caused by the sudden fault current and subsequent control action.
The TCS results show that the a.c. fault current flows for half a cycle after fault
initiation. During this period, the rectifier delay angle is retarded into the
inverting region (between 120 and 130 degrees) to accelerate the d.c. line
discharge. Since the fault is close to the rectifier, this effect is not observable on the
d.c. power trace of Fig. 13.21, but is clearly shown in the a.c. power response.
Once control action has isolated the d.c. line, fault arc extinction is dependent
on the natural oscillatory response of the d.c. line itself. This period combined
with that required for deionization gives a minimum shutdown time of 220 ms or
11 cycles.'' O )
At the end of the extinction and deionization period, transmission recovery
begins and in the absence of prior knowledge of the TCS response, the QSS model
produces instantaneous power recovery at this time, as illustrated by curve 1 of
Fig. 13.22.
 8.0

4.0

0.0

-4.0
-
-

-
-
D.C.-
L

-80 I I I I I I I I I
0.0 20 4.0 6.0 8.0 10.0 12.0 14.0 16.0 180 20.0
Time (cycles)

Fig. 13.21 . A.c. and d.c. apparent power at rectifier terminal for d.c. line fault

1
Start TCS

5.0
I End TCS

-
E
-3
, 3.0
4.0

- ( 1 ) QSS 1st run

( 2 ) QSS Delayed
( 3 1 QSS Matched
2.0-
( 4 ) TCS
1.0-

I
O?' 0.2 0.3 0.4 0.5 0.6
Time (seconds)

Fig. 13.22. Power flow to rectifier terminal for the d.c. line fault

However, as the TCS progresses, it becomes clear that transmission will not be
re-established at this time and matching between the two programs will be poor.
A number of subsequent equivalents may be obtained. The one adopted in this
case has the restart delayed and is ramped to nominal d.c. current linearly over a
two-cycle period. This equivalent provides excellent matching between the two
models as illustrated by curve 3 in Fig. 13.22.
As shown in Fig. 13.22, there is a 7 percent overshoot in the TCS link power at
full recovery. This is due to the finite rate of response of the controller. The
simulation is continued after recovery for a further four cycles to ensure that the
oscillation is damped and converges to the same conditions as the QSS model.
Throughout the d.c. fault there are no consequential converter disturbances
and because of this, and the accurate matching achieved with the QSS model, the
differences between the QSS and the TCS models are small.
 Start TCS End TCS

.-. 'V

(1 QSS Matched
0.9 ( 2 ) TCS
I I I I I I
0.8
0.1 0.2 0.3
Time (seconds)
0.4 0.5 0.6 *
(0)

A
Start TCS

-
2.0 -

1.0-
-. -. '1.\
1
!
( 1 QSS Matched
(2)TCS

n
a
67 I
0 I
I 0.2
I
1.0 - I
.A-
I /.'
1. (b)

Fig. 13.23. Voltage and reactive power at the rectifier for the d.c. line fault. (a) A.c.
terminal voltage; (b) Reactive power flow

Immediately after fault application, extra reactive power is required by the

rectifier. This is due to the half cycle of fault current which flows and the
associated commutation angle increase as shown in Fig. 13.23(b).This causes a
small depression of voltage at the rectifier terminals, which is illustrated in Fig.
13.23(a).
Once the fault current is cleared by control action, this depression changes to a
rise in voltage due to the extra reactive power available to the network from the
filters.
Towards the end of the fault period, the link draws reactive power from the
filters as the line is recharged. The oscillatory response after fault recovery rapidly
decays as control action settles. Differences at the inverter terminal are minor.
The difference in swing angles of the machines near the d.c. link terminals for
the two models is shown in Fig. 13.24. The results correspond to curves 1 and 4 of
Fig. 13.22.In this case, there is a considerable error in the TS assessment with a 16
percent error in the rectifier a.c. system machine first swing peak of the QSS
model.
 Fig. 13.24. Swing curves of Benmore (rectifier) and
Haywards (inverter) synchronous machines for the d.c. line
fault

Table 13.4. First swing maximum angles of Benrnore (rectifier)

and Haywards (Inverter) synchronous machines for the d.c. line
fault

First swing maximum angle

Benmore Haywards
(Rectifier) (Inverter)

QSS (1st run) 24.1" 45"

QSS (delayed) 29.1" 50.0"
QSS (matched) 29.0" 49.7"
TCS 28.7" 49.2"

However, as the matching between the QSS and TCS models is improved, the
differences in peak swing angle are reduced. For curve 2 of Fig. 13.22, the
difference is 2 percent and for curve 3 it is less than 1 percent. The maximum swing
angles for these cases are included in Table 13.4.These differences are particularly
small and are due in part to the improved matching between models and also to
the fact that the peak swing of both machines occurs very close to the d.c. fault
clearance time (0.43 sec. and 0.41 sec. respectively). Hence, any differences during
the link restart period, i.e. ramped or instantaneous in the case of the QSS model,
have little effect on the peak swing angle.

Conclusion
The three case studies presented in this section demonstrate the feasibility and
effectiveness of combining both the TCS and TS programmes. Together they
provide more accurate modelling of a.c.-d.c. systems disturbances than either
programme can provide on its own. Detailed information can be obtained from
the TCS regarding fault development, d.c. link response, converter performance,
over voltages, over currents, e t ~ . ( ~ l))" l
The different results obtained from the TS analysis of the three cases
investigated, indicates that a QSS model cannot be classified as giving either
pessimistic or optimistic results for peak swing angles. This emphasizes the need
to perform worst case transient converter simulations during investigations of a
system so that the QSS link model performance can be modified to more
accurately represent d.c. link performance for further studies. The system used for
these case studies is relatively strong and a weaker system, where stability is more
marginal, may produce much greater differences between the two models.

13.7. OVERALL CONCLUSIONS

In this final chapter, a compromise between the accurate but computationally
prohibitive, transient converter simulation and the computationally efficient, but
less detailed, transient stability program has been described. The advantage of
both these methods can be retained by simultaneously performing transient
stability and transient converter simulations with periodic coordination of the
results. In this way, the detailed converter model is provided with a time-variant
Thevenin equivalent for correct a.c. system representation as described in
Chapter 11. Similarly, from the transient converter simulation, accurate power
frequency information can be derived to represent the d.c. system behaviour at the
converter terminals. The computational requirements are minimized by restrict-
ing the use of the transient converter simulation to the most disturbed part of the
study. As the d.c. link regains normal controllability and attains predictable
behaviour, this detailed simulation is replaced by the quasi-steady-state model.
The results using the coordinated program have demonstrated its viability and
usefulness. The New Zealand primary transmission system used as a test, is a
strong system with few stability problems. The results clearly illustrate the need
for the more realistic modelling proposed in this chapter.

13.8. REFERENCES
1. W. Shepard and P. Zakikhani, 1972. 'Suggested definition of reactive power for
nonsinusoidal systems', Proc. IEE, 119 (9), 1 361 -1 362.
2. D. Sharon, 1973. 'Reactive power definitions and power factor improvement in
nonlinear systems', Proc. IEE, 120 (6), 704-706.
3. E. Micu, 1973. 'Suggested definition of reactive power for nonsinusoidal systems',
Proc. IEE, 120 (7), 796-797.
4. B. P. Lathi, 1968. Communication Systems. Wiley and Sons, New York.
5. W. T. Cochran, et al., 1977. 'What is the fast fourier transform? Proc. IEEE, 55 (lo),
1664- 1674.
6. F. J. Harris, 1978. 'On the use of windows for harmonic analysis with the discrete
Fourier transform', Proc. IEEE, 66 (1).
 7. K. S. Turner, M. D. Heffernan, C. P. Arnold and J. Arrillaga, 1981. 'Computation of
a.c-d.c. system disturbances-Part 11. Derivation of power frequency variables from
converter transient response.' IEEE Winter Power Meeting, Atlanta, Ga.
8. K. S. Turner, 1980. 'Transient stability analysis of integrated a.c. and d.c. power
systems.' Ph.D. thesis, University of Canterbury, New Zealand.
9. J. Arrillaga, M. D. Heffernan, C. B. Lake and C. P. Arnold, 1980. 'Fault studies in a.c.
systems interconnected by h.v.d.c. links', Proc. IEE, 127, Pt. C (I), 15-19.
10. M. D. Heffernan, K. S. Turner, J. Arrillaga and C. P. Arnold, 1981. 'Computation of
a.c-d.c. system disturbances. Part I-Interactive coordination of generator and
converter transient models.' IEEE Winter Power Meeting, Atlanta, Ga.
11. M. D. Heffernan, J. Arrillaga, K. S. Turner and C. P. Arnold, 1980. 'Recovery from
temporary h.v.d.c. line faults.' IEEE Summer Power Meeting, Minneapolis, Ma.
 Appendix I
Numerical. Integration Methods
1.1. INTRODUCTION
Basic to the computer modelling of power-system transients is the numerical
integration of the set of differential equations involved. Many books have been
written on the numerical solution of ordinary differential equations, but this
appendix is restricted to the techniques in common use for the dynamic
simulation of power system behaviour.
It is therefore appropriate to start by identifying and defining the properties
required from the numerical integration method in the context of power system
analysis.

1.2. PROPERTIES O F THE INTEGRATION METHODS

Accuracy
This property is limited by two main causes, i.e. round-off and truncation errors.
Round-off error occurs while performing artithmetic operations and is due to the
inability of the computer to represent numbers exactly. A word length of 48 bits is
normally sufficient for scientific work and is certainly acceptable for transient
stability analysis. When the stability studies are carried out on computers with a
32-bit word length, it is necessary to use double precision on certain areas of the
storage to maintain adequate accuracy.
The difference between the true and calculated results is mainly determined by
the truncation error, which is due to the numerical method chosen. The true
solution at any one point can be expressed as a Taylor series based on some initial
point and by substituting these into the formulae, the order of accuracy can be
found from the lowest power of step length (h) which has a nonzero coefficient. In
general terms, the truncation error T ( h )of a method using a step length h is given
by
T ( h )= O(hP )
+ ' (1.2.1)
where superscript p represents the order of accuracy of the method.
The true solution y(t,) at t, is thus
where y, is the value of y calculated by the method after n steps, and E, represents
other possible errors.

Stability
Two types of instability occur in the solution of ordinary differential equations,
i.e. inherent and induced instability.
Inherent instability occurs when, during a numerical step by step solution,
errors generated by any means (truncation or round-off) are magnified until the
true solution is swamped. Fortunately transient-stability studies are formulated
in such a manner that inherent instability is not a problem.
Induced instability is related to the method used in the numerical solution of
the ordinary differential equation. The numerical method gives a sequence of
approximations to the true solution and the stability of the method is basically a
measure of the difference between the approximate and true solutions as the
number of steps becomes large.
Consider the ordinary differential equation:

with the initial conditions y(0) =yo which has the solution

Note that R is the eigenvalue'') of the single variable system given by the ordinary
differential equation (1.2.3). This may be solved by a finite difference equation of
the general multistep form:

where ai and Bi are constants.

Letting
k
m(z) = 1(~~(2)~
i=O
and

and constraining the difference scheme to be stable when R =0, then the
remaining part of (1.2.5)is linear and the solutions are given by the roots zi (for
i = 1,2,. ..,k) of m(z) = 0. If the roots are all different, then

and the true solution in this case (A = 0) is given by

where superscript p is the order of accuracy.

 The principal root z,, in this case, is unity and instability occurs when lzil r 1
(for i = 2,3,. ..,k, i f 1) and the true solution will eventually be swamped by this
root as n increases.
If a method satisfies the above criteria, then it is said to be stable but the degree
of stability requires further consideration.
Weak stability occurs where a method can be defined by the above as being
stable, but because of the nature of the differential equation, the derivative part of
(1.2.5) gives one or more roots which are greater than or equal to unity. It has been
shown by Dalquist'') that a stable method which has the maximum order of
accuracy is always weakly stable. The maximum order or accuracy of a method is
+ +
either k 1 or k 2 depending on whether k is odd or even, respectively.
Partial stability occurs when the step length (h) is critical to the solution and is
particularly relevant when considering Runge-Kutta methods. In general, the
roots zi of (1.2.7) are dependent on the product h1 and also on equations (1.2.6).
The stability boundary is the value of h1 for which lzil = 1, and any method which
has this boundary is termed conditionally stable.
A method with an infinite stability boundary is known as A-stable (uncon-
ditionally stable). A linear multistep method is A-stable if all solutions of (1.2.5)
tend to zero as n + co when the method is applied with fixed h > 0 to (1.2.3) where
1is a complex constant with Re(1) < 0.
Dalquist has demonstrated that for a multistep method to be A-stable the
order of accuracy cannot exceed p = 2, and hence the maximum k is unity, that is,
a single-step method. Backward Euler and the trapezoidal method are A-stable,
single-step methods. Other methods not based upon the multistep principle may
be A-stable and also have high orders of accuracy. In this category are implicit
Runge-Kutta methods in which p < 2r, where r is the number of stages in the
method. '
A further definition of stability has been introduced re~ently,'~) i.e. Z-stability
which is the multivariable version of A-stability. The two are equivalent when the
method is linear but may not be equivalent otherwise. Backward Euler and the
trapezoidal method are Z-stable single-step methods.
The study of scalar ordinary differential equations of the form (1.2.3) is
sufficient for the assessment of stability in coupled equations, provided that 1are
the eigenvalues of the ordinary differential equations. Unfortunately, not all the
equations used in transient stability analysis are of this type.

A system of ordinary differential equations in which the ratio of the largest to the
smallest eigenvalue is very much greater than one, is usually referred to as being
stiff. Only during the initial period of the solution of the problem are the largest
negative eigenvalues significant, yet they must be accounted for during the whole
solution.
For methods which are conditionally stable, a very small step length must be
chosen to maintain stability. This makes the method very expensive in computing
time.
 The advantages of &stability thus become apparent for this type of problem as
the step length need not be adjusted to accommodate the smallest eigenvalues.
In an electrical power-system the differential equations which describe its
behaviour in a transient state, have greatly varying eigenvalues. The largest
negative eigenvalues are related to the network and the machine stators but these
are ignored by establishing algebraic equations to replace the differential
equations. The associated variables are then permitted to vary instantaneously.
However, the time constants of the remaining ordinary differential equations
are still sufficiently varied to give a large range of eigenvalues. It is therefore
important that if the fastest remaining transients are to be considered and not
ignored, as so often done in the past, a method must be adopted which keeps the
computation to a minimum.

I .3. PREDICTOR - CORRECTOR METHODS

These methods for the solution of the differential equation
pY=F(Y,X) with Y(0)=Yo (1.3.1)
and X(0) = X o
have all been developed from the general k-step finite difference equation:

Basically the methods consist of a pair of equations, one being explicit (BO= 0) to
give a prediction of the solution at tn+, and the other being implicit (Po # 0)
which corrects the predicted value. There are a great variety of methods available,
the choice being made by the requirements of the solution. It is usual for
simplicity to maintain a constant step length with these methods if k > 2.
Each application of a corrector method improves the accuracy of the method
by one order, up to a maximum given by the order of accuracy of the corrector.
Therefore, ifthe corrector is not to be iterated, it is usual to use a predictor with an
order of accuracy one less than that of the corrector. The predictor is thus not
essential, as the value at the previous step may be used as a first crude estimate,
but the number of iterations of the corrector may be large.
While, for accuracy, there is a fixed number of relevant iterations, it is desirable
for stability purposes to iterate to some predetermined level of convergence. The
characteristic root (2,) of a predictor or corrector when applied to the single
variable problem:
py = Ay with y(0) = yo (1.3.3)
may be found from:
 When applying a corrector to the problem defined by equation (1.3.3) and
rearranging equation (1.3.2) to give:

the solution to the problem becomes direct. The predictor is now not necessary as
the solution only requires information of y at the previous steps, i.e. at yn for
++

i = l,2, ...k.
Where the problem contains two variables, one nonintegrable, such that:
py = Ay + px with y(0) = yo (1.3.6)

then:

where

and

, ,
Although cn+ and mn+ are constant at a particular step, the solution is iterative
using equations (1.3.7) and (1.3.8). Strictly in this simple case, xn+, in equation
(1.3.8) could be removed using equation (1.3.7) but in the general multivariable
case this is not so.
The convergence of this method is now a function of the nonlinearity of the
system. Provided that the step length is sufficiently small, a simple Jacobi form of
iteration gives convergence in only a few iterations. It is equally possible to form a
Jacobian matrix and obtain a solution by a Newton iterative process, although
the storage necessary is much larger and as before, the step length must be
sufficiently small to ensure convergence.
For a multivariable system, equation (1.3.1) is coupled with:

and the solution of the integrable variables is given by the matrix equation
K + I = Cn+1+ M n + l . C Y n + d n + l l t (1.3.12)
,
The elements of the vector Cn+ are as given in equation (1.3.9) and the elements
of the sparse Mn+, matrix are given in equation (1.3.10).
The iterative solution may be started at any point in the loop, if Jacobi iteration
is used. Because the number of algebraic variables (X) associated with equations
(1.3.1) or (1.3.11) are small, it is most advantageous to extrapolate these algebraic
variables and commence with a solution using equation (1.3.1 1).
The disadvantage of any multi-step method (k > 2) is that is not self-starting.
After a discontinuity k - 1, steps must be performed by some other self-starting
method. Unfortunately, it is the period immediately after a step which is most
critical as the largest negative eigenvalues are significant. As k - 1 is usually
small, it is not essential to use an A-stable starting method. Accuracy over this
period is of more importance.

1.4. RUNGE -KUTTA METHODS

Runge-Kutta methods are able to achieve high accuracy while remaining single
step methods. This is obtained by making further evaluation of the functions
within the step. The general form of the equation is:

where

k,= hf(tn+cih,yn+ x
j= 1
aijkj) for i = 1,2,. . . , u (1.4.2)

Being single-step, these methods are self-starting and the step length need not be
constant. If j is restricted so that j < i, then the method is explicit and c, must be
zero. When j is permitted to exceed i, then the method is implict and an iterative
solution is necessary.
Also of interest are the forms developed by Merson and Scraton. These are
fourth-order methods ( p =4) but use five stages (u = 5). The extra degree of
freedom obtained is used to give an estimate of the local truncation error at that
step. This can be used to automatically control the step length.
Although they are accurate, the explicit Runge-Kutta methods are not A-
stable. Stability is achieved by ensuring that the step length does not become large
compared to any time constant. For a pth order explicit method the characteristic
root is:

where the second summation term exists only when v > p and where ai are
constant and dependent on the method.
For some implicit methods the characteristic root is equivalent to a Pade
approximant to eh'.
 The Pade approximant of a function f ( t ) is given by

and i f

then

If the approximant is to have accuracy of order M + N and i f f (0)= PMN(f (0))

then

It has been demonstrated that for approximants of Ah where M = N , M =

N +1 and M = N + 2, the modulus is less than unity and thus a method with a
characteristic root equivalent to these approximants is A-stable as well as having
an order of accuracy of M N .+
1.5. REFERENCES
1. L. Lapidus and J. H. Seinfeld, 1971. Numerical Solution of Ordinary Differential
Equations. Academic Press, New York.
2. G. Dalquist, 1963. 'Stability questions for some numerical methods for ordinary
differential equatios', Proc. Symposia in Applied Mathematics.
3. V. Zakian, 1975. 'Properties of I,, and J M N approximants and applications to
numerical inversion of laplace transforms and initial value problems', J. of Math.
Analysis and Applications, 5Q (I), 191-222.
 Appendix I1
Power-System Data
The programs described in the text have been applied to the New Zealand
Electricity power system. The primary a.c. transmission systems for the North
and South Islands are given in Figs. AII.l and AII.2.

Morsden

Bunnvthoroe II I
Brunsw~ck

New Plymouth Strotford

Hoywords

D.C link

Fig. AII.1. New Zealand North Island primary system

 Kikiwa

Smelter

Tiwai

Fig. AII.2. New Zealand South Island primary

system

The data presented here represents the input and output of a transient stability
study of the two primary a.c. systems plus the asynchronous h.v.d.c. in-
terconnection. The a.c. fault is in the South Island (rectifier) system and the
kv.d.c. link is assumed controllable throughout the study.
The loading data in the input is the solution from a load-flow program which
may be an integral part of the transient stability program or, as in this case, be run
separately. Also note that output of the AVR and speed governor information has
been restricted to two synchronous machines each.
Examples of the data as required for a short-circuit analysis program and for
various load-flow programs are given in the relevant chapters.
33333 >033) 13553 >0=30 a355
000J> 30391 03>03 07300 3303

...............
33333 32-35 33333

........................
>>>a> >a>>> 3>333
30031
133,)
0a331
3033
213) ZJ 3
3 3
1 >
-0331

.........................................
-73aU 30-)3
1>>31 33353
=333>
=>a33
3353>
a3333
1 > 3 3 1
3>33>
00
33

- -
C I 0 3 0 OOO=O 01331 00033 00300 00000 00303 00
33a23 ..91d2 0--39 33000 0030
a 3 31

U
d
3a033 30350 00333 030013 -000 Z 0 0 f C O G 0 tO\lU)O OZrOO Nr(?0O OU3-0 O-5-C NO
00300 00000 OOJOO OODOO ooon 4 - o o o o l c o o ~ o m o as o - r o ma-a- no o o o n m m o
33337 03300 0>300 00-oo 0000 9-3 3 o 7 1U3UO ? f a o r y w m n m -mr-v fa--= --==a -3

........................
3311'3
a3372
03a=o a3301
35333
z.aaaa
n033U
0030
3333
1. 3
&,1
...................................
3
o
1
CJ
a
~ ~ - nnr-n
0 3
2J-DU
n c r r a an*=-
17713
o . r a r a aaao3 oo
-3UUa O X 3 a 3 0=

-
53333 33303

-
J
33330 01333 00331 N300, 0003 Y) 0 0 0 03003 00033 0a000 03030 >5331 030- 30
U 0 00 0
Y 3
v n?
30, 3 0 3 3 3 n 3 3 3 3 33333 03533 03=03 00033 0 3 3 0 3 3 3 3 3 3 003-3 33333 0 3 3 0 3 3
onn - - ~ n - 3,-nn tr-17 ~ ~ a n> ~ t q t tOD>-- >>CVI n.-7, 3332- - 3 5 - 1 --u=n 7
U
t3-
>>a
on-
O
--a33 -----
-~ c u n na-n-- --t--
33333 3333-
OCOOO 0 0 0 0 ~ ooOo0
t t n o t U.-UU

0 0 0 ~ 0O O ~
0 0 3 0 0 n ~ q u f ir - - ~ -

...........................................................
171-1 ->3--? 11cc0 3 3 1 1 0 = o n > > ---DO
,-
U U % > ~0 - - 0 0
3 3 2 3 3 3 3 3 > 3 3 3 3 > 3 ---=a
~ o
00000 00000 00000
1177-
0--,3
>:>a-
00303
>ton- nnot-
a ~ n - - nrno?
-7333 33--3
03000 O O ~ O Oo
n
n
3
I

m
 I
I
I
I

. I .
IYI
I

.=;:
V) I 0
W 3 r
3 n l r o o ~ ~ o o o o l o o m m ~ o o 3 a O9.-
m
I1
I
A, I

-.
0

3
1
I
1I
0
11
W
5
aaYI
x2

.IC
P-J 2xxxsssXLsauaaauuQ.auaau
DL
"la
32
0
X d
OX
X 3
- 2Z
W
>W
% II f C
",<a
XI.
4x2

a I
3 I
.r
LL I
I

3 :I -.
U
" I P
-a
v
II
~
W
n
N
I Il i u)
I I
I I
J I

-:I
V
rl II
I
>
*
- 1
I 3.

-
t
.
l
I I
Z
I I:
I W
2 1 -
3 1 -
4 1 -
P I " l
 I
I
I
I
I .

.
I "0
I
" I *
a
0
:1
'-!
0
t
N
. lI
I
U

a l e
a
'Y
1
1
02 11 -

4L 1I .

-' II "
W I U
-41 II u
U
I I
I I
W.,
J I
II D
- 1 7 .

--
3 1
-&- 1I .
1 3
c 1 z
I
II !,
2 1 -
3 1 0
- 1 -
r l 'A
 -
3 c

--. -
3
I- -a
-
0

rc
3.-
0

0
a0
J3
%
111-
Id

-4
a", 3 m
-aC3 u
J
3
J a
3-
1
x-z
X J
*
C
I
c
z
x
J
-
J
-LI --
Z
A!

**J
3.72
0
I+
xc
J)

L
UJ 30

- .. -.
4 4
SZ
23
-0
00

00
An
h.

I
I
w-
>x
J2
4 ..
L

-
-
-J
A 4
J3
5 .
L
00
33

-.-.
f<
U
\23
4-
z
rn 3

-
- n..>
m N
Y . u-
T
3 t N
A! J
C >J

- ..
JU >- JZ
>.

-.
-L > 0 -7
L9
% J)
00 21
3

2 1
-
3
un

m-t

om
dc
JJ 3.5
1- 3.0
1 93.
m
J
a
..
31

,>
01
T
C
..
a-
3.-
03
c-
3, ..
01
-a
00

z J)
un
'3 3 2% 00
7- 1 I
w"3
35 J"l
u
.o .
T2
4-
0-
un
..
OD
nu7

-
-Q 0I %
I
-4

1 I u
.3 .
.7 I 1
.*
9Z W
I.,I LL
"34. 3 r Q
.-

-
3 2 S3 44
.U WT
I I
L 1: I 1
4 n
31d D I fd. d
*PA! l
1114
32
0

a
-
o x
I 3
AJr
Wn
z
Y

I I

'1111
>d

.*
C I
d3 O-Z
0
0-
n
-,-I
C U
-l ax
u
WC
-,
u x
31d at
m
c1Q.O-
rl-o-
I
:!
:
:I
50'0-
on-o-
 Subject Index
A-stable, 23 1,397 Characteristic root, 23 1,398
A.C. faults, 180, 184, 189,228, 347 Cornmutating voltage, 46, 129
A.C. load flow, 58.90 Commutation, 45, 129
angle reference, 59 abnormal, 307
balanced, 58 failures, 387
power mismatches, 65, 7 1,94 normal, 307,309
unbalanced, 9 1 Commutation angle, 46
voltage reference, 59 Commutation reactance, 46, 50
A.C. system model Compound turbine, 261
impedance angle, 328 Connection matrices, 6, 13
impedance frequency locus, 327,347 compound matrices, 15
Thevenin equivalent, 282,328,338, three phase transformers, 28
345,347 Constant current control, 53, 277,3 18,
time-variant, 328,338,367 354
A.C.-D.C. conversion, 43,271 Constant extinction angle control, 53,3 18,
A.C.-D.C. load flow, 124, 152 359
control equations, 129, 159 Constant power control, 55,277
converter variables, 126, 156 Contactors, 293
mismatch equations, 125 Convergence, 8 1, 14 1,229
per unit system, 128 Converter
sequential algorithm, 124, 137 commutation (source) voltage, 46
unbalanced, 153 commutation reactance, 46,50
unified (simultaneous) algorithm, 124, controlled rectification, 45
13 1 faults, 344
Admittance matrices, 5 inversion, 48
formation for simple networks, 17 modes of operation, 273,307
formation for three phase networks, 4 1 parallel connection, 5 1
three phase compound, 13.21 phase angle control, 152
Angular momentum, 208 series connection, 50
Applicability of A.C.-D.C. models, 381 six-pulse, 309
Automatic voltage regulator, 95,214,238 transient model, 306
twelve pulse. 309
Backward Euler method, 233 uncontrolled rectification, 44
Boiler, 2 18 Converter control
Branch switching, 228 characteristics, 54,272
constant current, 53,277,3 18,354
Cage factor, 268 constant power, 55,277
 firing instant, 3 18, 363 interactive coordination, 336,356, 382
firing limits, 3 18 nodal analysis, 30 1, 345
power modulation, 279 partitioning, 302,304
Converter transformer, 44, 306 state variables, 330, 335
connection, 162, 1 74 topological matrices, 30 1
leakage reactance, 44,307,3 1 1,323 valve voltages and currents, 3 13
magnetization curve, 324 variable topology, 303
magnetizing reactance, 324 voltage and current relations, 302,3 13
on-load tap changer, 44,48, 127, 155, Dynamic stability, 206, 26 1
323
phase-shifting, 43 Eigenvalue, 234,397
saturation, 324 Enforced delay, 274
tertiarv winding, 5 1 Equidistant firing control, 152, 159,3 17
transient model, 323 Exciter, 2 16, 240
Cylindrical rotor, 2 10
Fast decoupled load flow
D.C. faults, 353, 360, 364, 389 assumptions, 80,98
arc extinction and deionization, 357, balanced A.C. system, 78
360,389 balanced A.C.-D.C. system, 13 1
detection, 354, 356, 360 characteristics and performance, 83,
earth resistance, 358,36 1 105, 144, 167
restart, 359, 362, 368 flow diagram, 8 1, 101, 135, 165
D.C. transmission, 2, 53, 124, 154, 168. multiterminal D.C., 139
32 1 sequential solution technique, 137
monopole, 329 simultaneous solution technique, 131
multiterminal, 57, 139 unbalanced A.C.-D.C. system, 152
New Zealand link, 5 1,329, 357 unbalanced A.C. system, 96
n-equivalent, 329 unbalanced system program structure,
quasi-steady state, 277, 337,345, 382, 103, 166
389 unified solution technique, 13 1
steady state, 43 Fault calculations, 184, 189
submarine cable, 329,357 Fault studies, 180
two-terminal, 53 Filters, 46, 5 1, 155,305,326,367
Dead band, 286 Fourier analysis, 47
Decoupled load flow methods, 76 fundamental current, 47, 160
Deep bar rotor, 267 spectral analysis, 370, 372
Delay angle, 45, 53, l52,27 1 spectral leakage, 373
Direct axis, 207 TCS waveforms, 3 7 1
Direct current, 44, 53 Frames of reference, 207,224,267
margin, $5
ripple, 44 Harmonics, 156, 177,353,367,370,374
setting, 55 Hydrogovernor, 2 17
Distance relays, 293 Hydroturbine, 2 18,26 1
Double cage rotor, 267
Dummy busbar, 229 Induction machine, 265
Dynamic A.C.-D.C. model, 300 cage factor, 268
A.C. faults (see A.C. faults) contactors, 293
branch equations, 302 deep bar rotor, 267
computer program, 332, 340 double cage rotor, 267
D.C. faults (see D.C. faults) inertia constant, 266
dependent variables, 303,335 magnetizing reactance, 267
initial conditions, 3 15 mechanical torque, 266
Induction machine (contd.) Nodal formulation
open circuit reactance, 267 dynamic A.C.-D.C. model, 300
slip, 265 load flow, 60
transient reactance, 26 7 transient stability, 223
Inertia constant, 208,266 Nonunit protection, 229
Infinite machine, 214 Norton equivalent, 63,224
Integration, 230,330, 395 Numerical stability, 33 1,396
A-stable, 23 1,397
backward Euler method, 233 Open circuit faults, 192
characteristic root, 23 1, 398 Open circuit reactance, 267
implicit, 33 1 Ordering of sparse matrices
Runge-Kutta methods, 230, 331,400 dynamic, 74
X-stable, 23 1, 397 pre-ordering, 74
step length, 23 1,33 1 Overcurrent relays, 290
trapezoidal method, 23 1
Interactive coordination, 336,356,382 Per unit for D.C. system, 48, 128, 157,33 1
Interceptor governor, 265 Potier reactance, 252,257
Interceptor valve, 26 1 Power factor in A.C.-D.C. systems, 49
Inversion, 48 Power-frequency control, 56
characteristic, 54 Predictor-corrector methods, 398
deionization angle, 48 Primitive networks. 5
extinction angle, 48,53
Iteration schemes Quadrature axis, 207
A.C. load flow, 72,8 1, 101
A.C.-D.C. load flow, 132,165 Reactive power in A.C.-D.C. systems, 50,
transient stability, 247 53
Rectification, 44,27 1
Jacobian matrix elements abnormal modes of operation, 273
A.C. load flow, 68 characteristics, 54,272
A.C.-D.C. load flow, 132, 167 controlled, 45
delay angle, 45,53,271
Lead-lag circuit model, 2 19 dynamic load, 273
Linear transformation, 4, 15,29 enforced delay, 274
Load characteristics, 220 operating mode identification, 276
Loads, 64,220,326 saturable reactors, 27 1
static load, 272
Machine switching, 230 uncontrolled, 45
Magnetizing reactance, 267 Relays, 290
Matrix partitioning, 14,24 distance, 293
Matrix sparsity, 60 induction motor contactors, 293
Mechanical power, 208,218 overcurrent, 290
Mechanical torque, 266 undervoltage, 293
Resonance, 354
Network subdivision, 18 Runge-Kutta methods, 230,33 1,400
New Zealand Electricity power system,
329,402 S.C.R. (see Short circuit ratio)
Newton-Raphson S.V.S., 286
characteristics and performance, 75 deadband, 296
equations, 67 Saliency, 2 10,225
flow diagram, 72 Salient pole rotor, 2 10
general formulation, 66 Saturable reactors, 271
Jacobian elements, 68 Saturation, 251,307,324,353
starting techniques, 75 Saturation curve, 253,257
Saturation factor, 252 Synchronous machine models. 208.21,4.
Sequence components (see Symmetrical 237
components) Synchronous reactance. 2 10
Series element three-phase representation. System damping. 279
11.26
Short circuit faults, 189 TCS. 300
Short circuit ratio, 354 alignment with TS, 377, 383
Shunt element three-phase representation, fault application and removal. 347. 378
25 interaction with TS, 336.369,376
C-stable. 23 1. 397 synchronous machine model. 3 19
Slip. 265 transformer model, 323
Smoothing reactor, 44,308 Thermal turbine, 2 18
Sparse matrix equations, 74, 167, 184, 303, Thevenin equivalent. 223,345
340 Three phase faults. 18 1. 228
Sparsity, 184, 303 Three phase representation, 1 1, 18.26
ordering, 74 Transformer model
programming, 72, 167 saturation, 324
Speed governor, 2 17 single phase, 7
interceptor. 265 three phase, 26, 162
Stability, 206, 25 1 (see also Converter transformer)
numerical, 396 Transient converter simulation (see TCS)
representation of plant in network, 224, Transient reactance, 2 1 1, 267
270,28 1,289,295 Transient stability, 206
Starting techniques for load flow, 75 Transient stability program structure, 240
Static VAR compensation systems (see Transmission line model
S.V.S.) ABCD parameters. 39
Stiffness, 207, 397 line sectionalization. 39
Subtransient reactance, 2 12,269 long lines, 22
Swing curves. 388 mutual coupling, 22
Symmetrical components, 10, 18 1, 188,296 single phase, 6
three phase transformers, 35, 1 76 three phase, 18
Symmetrical firing control (see Equidistant Trapezoidal method, 23 1
firing control) Triangular factorization, 73, 136
Synchronous machine Turbine
angular momentum, 208 compound, 26 1
cylindrical rotor, 209 hydro, 2 18,26 1
damper windings, 208,2 12 thermal. 2 18,26 1
damping coefficient, 209,235
inertia constant, 208
negative sequence braking, 296 Unbalance
negative sequence impedance, 296 effect on converters, 152
Potier reactance, 252,257 source of unbalance, 90
salient pole rotor, 2 10 unsymmetrical A.C. system, 12
saturation, 25 1 Unbalanced faults, 187,295
saturation factor, 252 Undervoltage relays, 293
subtransient reactance, 2 12
transient reactance, 2 1 1 Valve
Synchronous machine controllers, 2 14, extinction, 3 18,332
238,265 switching, 33 1