IEEE TRANSACTIONS ON POWER SYSTEMS 1

Formulation of Three-Phase State Estimation

Problem Using a Virtual Reference
Andre L. Langner, Student Member, IEEE, and Ali Abur Fellow, IEEE

Abstract—State Estimation (SE) is the backbone of modern used. The modeling of three-phase distribution line segments,
Energy Management System (EMS) due its capability of pro- transformers, capacitors, voltage regulators, loads is detailed
cessing real time measurements and provide reliable information in [6], [7]. Special attention is given to the modeling of three-
to system operators. Since its introduction to power systems in
the 70’s, SE has been widely used in transmission networks. phase co-generators in [8]. The modeling of open wye-open
Distribution grids on the other hand lack sufficient number and open delta-open transformers is shown in [9], and general
of real time measurements, and for that reason SE has not transformer models are discussed in [10], [11].
been widely implemented on these systems. The recent increase Motivated by the growing interest in Distribution Manage-
in the number of renewable energy sources connected to the ment Systems (DMS), various investigations are initiated to
grid at lower voltage levels, the advent of Distribution Au-
tomation (DA) and Smart Grids necessitate closer monitoring develop a Distribution System State Estimator (DSE) in the
of distribution networks and thus forcing utilities to upgrade 90’s. One of the first results on this topic appeared in [5],
their operations and deploy Advanced Distribution Management where a state estimator was developed using a minimum num-
Sytems (ADMS). Therefore, Distribution System State Estimation ber of remote measurements. The proposed method requires
(DSE) is paramount to provide real time monitoring of active network reconfiguration information, remote measurements of
distribution grids. This papers investigates the formulation of
the three-phase distribution state estimation where in the most voltage, power, and current, along with statistical information
general case, none of the system buses may have balanced three pertaining to loads.
phase voltages. In such a case, the choice and definition of This was followed by numerous publications that used tra-
the slack bus need to be revisited. Most of the existing work ditional Weighted Least Squares (WLS) method [12]–[16] as
arbitrarily assign the root bus of the distribution feeder as well as more robust alternatives to cope with the large number
the balanced three-phase reference which may not be realistic
for today’s active distribution systems. This paper addresses of zero injection measurements by treating them as equality
this shortcoming of the existing SE formulations by introducing constraints [17], or using orthogonal transformations [18].
a virtual reference bus. The proposed approach is effectively Reference [13] discloses important aspects in the jacobian
validated by simulations carried out on distribution test systems. matrix due the coupling between phases and the high r/x
Keywords—State Estimation; Distribution systems; Three- ration. It develops a way of simplifying the jacobian by using
phase modeling; Reference Bus. sequence transformation to produce small phase voltages (less
than 10 degrees). It also reports convergence problems in the
I. I NTRODUCTION presence of current magnitude measurements.
In modern Energy Management Systems (EMS), State Es- Among these references, [14] highlights the common imbal-
timation (SE) acquires and processes raw data from remote ances in the three-phase network models, as well as addressing
terminal units (RTU) and phasor data concentrators (PDC), to related issues of lines in different voltage levels using the same
provide a real time snapshot of the current operating state of right of way, having non-grounded transformers at distribution
the entire transmission network. The estimated states are then level. It also includes generator models that use balanced
used by various network applications, such as optimal power excitation voltages behind machine reactances. The impact
flow, contingency analysis, transmission forecast, voltage sta- of not properly modeling untransposed lines and ignoring
bility assessment, just to name a few. unbalanced loads are illustrated in [19], implying that the ac-
The traditional formulation of the SE problem is based on curate representation of the multi-phase distribution systems is
the premise that transmission lines are fully transposed, and paramount in developing reliable and unbiased state estimators
loads are balanced, which allows the use of only the positive to monitor these systems. Reference [20] unveils a multiphase
sequence components [1] of the voltages and currents. Hence, power flow and state estimation based on asymmetric models
transmission system is modeled and analyzed as if it is a single of distribution networks and synchronized measurements.
phase network [2]–[4]. The scarcity of power measurements motivated the de-
The distribution system is typically identified as part of velopment of a branch-current based DSE [21], allowing
the utility system that is located between the distribution the decoupling of phases where the states are represented
substations and low voltage buses of distribution transformers in rectangular coordinates. Several other investigators made
[5]. Such systems are characterized as being unbalanced, valuable contributions to the branch-current approach in order
where loads are not evenly distributed among the phases, to further improve its performance [15], [22]–[25].
some line sections having single or two-phase configurations Unlike transmission networks, which are well covered with
or having lines which are not fully transposed. Thus, a multi- real-time measurements, distribution networks were always
phase detailed model of the entire system will have to be characterized by the lack of real time measurements for
IEEE TRANSACTIONS ON POWER SYSTEMS 2

economical reasons. Thus, many of the first attempts in trying timation of multi-phase systems even when they are operating
to deploy a DSE made use of historical load information [5], as a micro-grid and under highly unbalanced conditions. The
[13], [15] and stochastic loads models [16], [26] as pseudo- proposed method is first derived in detail and then validated
measurements to render the entire feeder observable. Meter using simulations carried out on IEEE test feeders with very
placement techniques were also developed to improve the different sizes. The results obtained using the existing methods
performance of DSE [23], [27]–[29]. Nowadays, Advanced are also shown to comparatively highlight the improvements.
Measurement Infrastructure (AMI), Automatic Meter Reading
(AMR) and Smart Meters are providing new sources of mea- II. E QUALITY C ONSTRAINED S TATE E STIMATION
surements for DSE, which have been exploited in [30]–[33]. In state estimation, measurements are modeled by a set of
The use of synchronized phasor measurements and µPMUs non-linear equations as a function of the states and the errors,
can also benefit DSE, as reported in [34]–[38]. State-of-the- as follows:
art reviews on the use of DSE are presented in [39], [40].
The increasing number of renewable generation, such as z = h(x) + ε (1)
solar PVs, in lower voltage levels also pose new challenges
for distribution operation. During some periods of the day, where z is (m × 1) measurement vector, x is the (n × 1) true
where the load consumption is low and solar generation is state vector, h(.) is the nonlinear function of measurements,
at its high, the power flow direction may revert. In this ε is the (m × 1) measurement error vector, m and n are the
case, the feeder do not operate in a passive way by only number of measurements and states, respectively.
demanding power, but it can act by actively injecting power Although the measurement error vector ε is unknown, it is
into the sub-transmission grid. Distributed Generation (DG) is assumed to have a zero mean Gaussian distribution. They are
changing the way distribution networks are operated, and its modeled using the variance σi2 associated with the measure-
impact on DSE are investigated in [41]–[43]. Reference [42] ment devices’ precision. The variances of all measurements
extends the state estimation method to identify unexpected are used to compactly form the covariance matrix given by:
invariable power injections from DGs at points of common
coupling (PCC). The authors use the Hachtel’s method, or R = E{T } = diagonal{σ12 , σ22 , ..., σm
2
} (2)
sparse tableua formulation, by dealing with null injections
The optimal state x̂ is found by minimizing the weighted
as equality constraints. Unexpected power injection at PCCs
sum of the residuals:
are identified by using normalized Lagrange multipliers and
geometric tests.
m
This paper is mainly concerned about the choice or formu- 1X 1
min J(x̂) = [zi −hi (x̂)]2 = [z−h(x̂)]T R−1 [z−h(x̂)]
lation of the reference phase angles in solving multi-phase x 2 i=1 σi2
unbalanced power system state estimation problem. While (3)
the specific solution method chosen to estimate the system This minimization problem turns out to be a constrained
state is not directly relevant to the presented formulation, minimization problem due the large presence of null injection
without loss of generality the well-known and commonly virtual measurements, and the problem can be reinstated as
used WLS method is used in the simulations. Furthermore, given in Eq. 4.
in order to avoid possible ill-conditioning problems due the
large number of zero injection measurements in distribution 1 T
min J= r Wr
networks, Hachtel’s augmented matrix approach is used in the 2
s.t. c(x̂) = 0 (4)
DSE implementation.
The main contribution of the paper is related to the choice r = z − h(x̂)
and modeling of the reference bus. When using Supervisory
where W = R−1 , c(.) is the non-linear function of null
Control and Data Acquisition (SCADA) measurements, it is
injection virtual measurements, and r is the vector of residuals
necessary to define a reference angle in order to consistently
given by the difference between the vector of measurements
express the measurement equations. In the case of three-phase
and the vector of estimated measurements.
networks, there is not much discussion in the literature about
The resulting Lagrangian will have two sets of multipliers:
the choice of a reference for three-phase state estimation
algorithms, and to the best of authors’ knowledge most of
L = J(x̂) + λT c(x̂) + µT (r − z + h(x̂)) (5)
the methods implicitly assume that the root bus of the radial
distribution feeder has a balanced three-phase voltage which and the optimality conditions will be given by:
can be used as the reference.
Relaxing this assumption will require performing state es- ∂L(x̂)/∂x = 0 ⇒ C T λ + H T µ = 0
timation using only one phase of this root bus as a reference. ∂L(x̂)/∂λ = 0 ⇒ c(x̂) = 0
However, as will be illustrated in the sequel, such an approach (6)
∂L(x̂)/∂r = 0 ⇒ W r − µ = 0
may lead to convergence problems. Hence, the main goal of
∂L(x̂)/∂µ = 0 ⇒ r − z − h(x̂) = 0
this paper is to develop and validate a new method where a
three-phase balanced reference is created using a virtual bus The third equation allows to eliminate r or µ (r = Rµ,
which does not impact the final solution but facilitates state es- r being the measurement residual vector) and by linearizing
IEEE TRANSACTIONS ON POWER SYSTEMS 3

the remaining three equations allows to form the system of where θnph and Vnph refer to voltage angles and magnitudes,
equation given in Eq. 7. respectively, at bus n for phase ph. Phase ph may be a, b, or
     c.
R H 0 µ ∆z k
     In the same way as positive sequence modeling, the mea-
H T 0 C T  ∆x̂ =  0  (7)
     surements are functions of the state x. However, the expres-
k sions for power flows and injections are generalized to account
0 C 0 λ −c(x̂ )
for three phase components, and the jacobian is augmented as
This is refereed to Hatchel’s sparse tableau algorithm [44], well.
and iteratively solving (7) yields an optimal solution avoiding
numerical problems.

III. M ODELING OF D ISTRIBUTION S YSTEMS

This section will only review the distribution system line
modeling and will refer the reader to [11] and OpenDSS A. Measurement Equations
manual for modeling details of transformer, voltage regulator
and capacitor banks due to lack of space.
Modeling of distribution line segments is critical in the Assuming the network topology is correct, and by consid-
feeder analysis [7] where these segments may be three-phase, ering the connectivity of the three-phase lines, transformers,
two-phase, single-phase, overhead or underground. They are and capacitors, along with their primitive matrices, the system
best modeled via impedance and admittance matrices. The admittance matrix Ybus will be formed as follows:
series impedance of a distribution line consist of the conduc-
tors resistance and the self and mutual inductive reactances Ybus = AT (Zprim )−1 A (10)
resulting from the magnetic fields surrounding the conductors.
The resistance component is found in conductors tables, and where A is the branch-node incidence matrix, and Zprim
the self and mutual impedances are calculated following the is system primitive impedance matrix which is formed by
Modified Carson’s equations as given in [7]. diagonally grouping the primitive matrices Zabc .
For most applications, Kron reduction will transform the Expressions for different types of measurements can be
primitive 4 × 4 impedance matrix (Zprimitive ) into a 3 × 3 written either in rectangular or polar coordinates. The equa-
one (Zabc ), consisting of the self and mutual equivalent tions used to represent the bus power injection and line flow
impedances for the three phases, taking neutral and/or ground measurements are described next. Voltage magnitudes are ex-
conductors out of equation. For an untransposed distribution plicitly described here since they directly measure a state to be
line, the diagonal terms will not be equal to each other as estimated (if the states are in polar form). Current magnitude
well as the off-diagonal terms. However, the resulting matrix measurements are not considered in this implementation. The
will be symmetrical [7]. Which means, Zaa 6= Zbb 6= Zcc and measurements are modeled using the states x (i.e. Via,b,c and
Zab 6= Zbc 6= Zac in (8). θia,b,c , i = 1, 2, ..., n), and the values from the Ybus matrix.
The active and reactive power injection at bus k phase ph can
be written as given in Eq. 11.
 
Z Zab Zac
 aa 
[Zabc ] = 
 Zba Zbb Zbc 
 (8) n
Zca Zcb Zcc
XX
Pkph = Vkph Vmp [Gph,p ph p
km cos(θk − θm )+
Ω/mile p∈Ω m=1

The shunt admittance of a line is composed of the conduc- Bkm sin(θkph − θm

ph,p p
)]
n (11)
tance (generally small and disregarded) and its susceptance. X X
Since distribution lines are short, the line charging is small, Qph
k = Vkph Vmp [Gph,p ph
km sin(θk − p
θm )−
p∈Ω m=1
around 4-6 µSiemens per mile, and can be neglected. However,
ph,p
when underground lines are considered, the line charging Bkm cos(θkph − θmp
)]
increases to 88-96 µSiemens per mile, causing significant
changes in the analysis [45]. Where Vkph is the voltage magnitude at bus k phase ph, Vmp
is the voltage magnitude at bus m at phase p, which belongs
IV. T HREE -P HASE S TATE E STIMATION to the set Ω composed by phases a, b, and c. θkph is voltage
p
angle at bus k phase ph, θm is the voltage angle at bus m
The same constrained WLS approach presented in section ph,p ph,p
phase p. Gkm and Bkm are the real and imaginary parts of
II applies to three-phase state estimation, but now the state
(k,m) element of the system Ybus matrix between phases ph
vector x is composed of voltage phasors for each phase, as
and p.
given below:
The active and reactive power flow from bus k to bus m
x = [θ1ph ...θnph V1ph ...Vnph ] (9) can be written as given in Eq. 12.
IEEE TRANSACTIONS ON POWER SYSTEMS 4

ph
Pkm =
ph
X
Vk Vmp [(Gph,p ph p ph,p ph p
km cos(θk − θm ) + Bkm sin(θk − θm ))]
p∈Ω
X ph,p
− Vkph Vkp [(Gph,p ph p 0 ph p
km cos(θk − θk ) + B km sin(θk − θk ))]
p∈Ω
pj
Qkm =
X
Vkph Vmp [(Gph,p ph
km sin(θk
p
− θm ph,p
) − Bkm cos(θkph − θm
p
))]
p∈Ω
Fig. 1: 2-Bus system
ph
X ph,p
− Vk Vkp [(Gph,p ph p
km sin(θk − θk ) − B 0km cos(θkph − θkp ))]
p∈Ω solutions for the state estimation problem. Hence, this paper’s
(12)
ph,p ph,p ph,p main focus is to develop a simple yet general approach
where B 0km = Bkm sh
+ Bkm sh
. Bkm is the line charging
which can overcome this limitation and can be used for any
susceptance between phases ph and p.
multi-phase state estimation solution operating under balanced
or unbalanced operating conditions. In order to contrast the
B. Jacobian Matrix developed approach with the two existing practices, these three
The jacobian consists of taking the partial derivatives of approaches will be comparatively presented and discussed in
power injection, flows, and voltage magnitude measurements sequence. The first approach assumes the existence of at least
with respect to the systme states, as shown in Eq. 13. one bus with balanced voltages and selects that bus as a three-
 a,b,c a,b,c a,b,c a,b,c 
phase reference. The second approach is similar to the first,
∂Pk ∂Pk ∂Pk ∂Pk
a,b,c a,b,c but it selects only one phase of a bus voltage as the reference
 ∂θka,b,c ∂θm ∂Vka,b,c a,b,c
∂Vm
and tries to estimate the remaining phase angles as well as the

 ∂Qk ∂Qa,b,c
k ∂Qa,b,c
k ∂Qa,b,c
k

rest of the system states. The third approach is the proposed
 a,b,c a,b,c
 ∂θk ∂θm ∂Vka,b,c a,b,c 
∂Vm 
 ∂P a,b,c a,b,c
∂Pkm a,b,c
∂Pkm a,b,c 
∂Pkm
 km  by this paper and will be explained below.
 a,b,c a,b,c
∂Vka,b,c a,b,c 
H =  ∂θka,b,c ∂θm ∂Vm
(13)
∂Qa,b,c ∂Qa,b,c ∂Qa,b,c

 ∂Qkm km km km 
 ∂θa,b,c a,b,c
∂θm ∂Vka,b,c a,b,c 
∂Vm A. Approach I - Assuming an existing bus with balanced three-
 k 
∂Vka,b,c phase voltage
 0 0 0 
 
 ∂Vka,b,c 
∂Vm a,b,c Most three-phase state estimation programs assign the sub-
0 0 0 a,b,c
∂Vm station bus (feeder head) to act as the reference by fixing the
Matrix C from the system in (7) will be similar to the a
angles of its three phase voltages, i.e. θsubs = 0◦ , θsubs
b
=
◦ c ◦
jacobian matrix H, expect that only power injection at buses −120 , θsubs = 120 . However, such an assumption may
without load and generation are taken into account, as given not always be valid since today’s distribution networks may
in 14. contain distributed energy sources, unbalanced loads which
 a,b,c a,b,c a,b,c a,b,c
 will force almost all bus voltages to be unbalanced including
∂Pj ∂Pj ∂Pj ∂Pj
the feeder head.
 ∂θa,b,c a,b,c
∂θm ∂Vja,b,c a,b,c
∂Vm
Consider the simple 2-bus system shown in Fig. 1. The
C =  ∂Qja,b,c (14)

j ∂Qa,b,c
j ∂Qa,b,c
j ∂Qa,b,c
j

∂θja,b,c a,b,c
∂θm ∂Vja,b,c a,b,c
∂Vm
voltage phase angles at bus 1 are assumed balanced and fixed
as the three-phase reference. The voltage angles at bus 2
where j stand for a null injection bus. are estimated, and the voltage magnitudes at both buses are
estimated using the available line model and the associated
V. R EFERENCE B US measurements. The line is assumed to be a untransposed
Unlike the case of three-phase power flow solution [46], distribution feeder with different self Zaa , Zbb , and Zcc and
[47], in solving the state estimation problem a reference mutual Zab , Zac , and Zbc impedances. The load connected at
bus can be assigned in an arbitrary fashion since the goal bus 2 is highly unbalanced.
is to select a reference phasor with respect to which all This system is modeled using the software OpenDSS [48],
other bus voltages can be defined [12]. Direct extension which solves the unbalanced distribution power flow and
of the positive sequence state estimation formulation to the provides the true states and associated error-free measure-
three-phase state estimation problem implies identifying and ments. Here, the source bus is modeled using small values
selecting a balanced three-phase bus voltage in the system. of short circuit power (200 MVA and 210 MVA of three and
However, given today’s active distribution systems such a bus single phase short circuit powers) to force highly unbalanced
may not exist in the entire distribution system, i.e. all bus solutions.
voltages including the commonly assumed reference bus at Power injection and voltage magnitude measurements are
the feeder head may have unbalanced voltages. In this most placed at both buses, and power flows in all phases to ensure
general case, the existing conventions of arbitrarily assigning a the system is observable, and all of them are assumed to have
balanced bus as the reference will yield biased and/or incorrect the same standard deviation of 1×10−3 . Perfect measurements
IEEE TRANSACTIONS ON POWER SYSTEMS 5

TABLE I: 2-Bus system states - Approach I 1) Injection Equations: The derivatives of power injection
measurements with respect to θ:
Busphase True States Estimated States

1a 0.9972 0.0000 0.9972 0.0000 ∂Pkph

= − Qph ph 2 ph
k − (Vk ) Bkk (15)
1b 0.9985 -119.8828 0.9985 -120.0000 ∂θkph
1c 0.9994 120.1602 0.9994 120.0000 ∂Pkph
= Vkph Vkp (Gph,p ph,p ph,p ph,p
kk sinθkk − Bkk cosθkk ), ph 6= p
2a 0.9934 -0.1927 0.9934 -0.1927 ∂θkp
2b 0.9995 -119.9997 0.9995 -120.1166 (16)

2c 0.9988 120.2226 0.9998 120.0624 ∂Pkph ph p ph,p ph,p ph,p ph,p

p = Vk Vm (Gkm sinθkm − Bkm cosθkm ), k 6= m
∂θm
(17)
are assumed, i.e. the results from power flow solution are used ∂Qph
k
= Pkph − (Vkph )2 Gph
kk (18)
directly without adding any Gaussian noise. The measurement ∂θkph
jacobian for this case will have 24 rows (3 active and 3 ∂Qph
reactive power flows, 6 active and 6 reactive active power k
= − Vkph Vkp (Gph,p ph,p ph,p ph,p
kk cosθkk + Bkk sinθkk ), ph 6= p
∂θkp
injections, and 6 voltage magnitude measurements) and 12
(19)
columns (corresponding to system states). Eliminating the
three columns related to voltage angles at bus 1, a full rank ∂Qph
k ph p ph,p ph,p ph,p ph,p
p = − Vk Vm (Gkm cosθkm + Bkm sinθkm ), k 6= m
jacobian H matrix is obtained. This implies that the system ∂θm
is observable and solvable. Starting from the flat start, a (20)
converged solution is found using a convergence tolerance
of 1 × 10−5 on the state updates. The weighted sum of At flat start, Vka,b,c and Vma,b,c are 1 p.u., θka = θm
a
= 0,
◦ ◦
the residuals (J-index or objective function), which is the θkb b c c
= θm = −120 , and θk = θm = 120 . As a conse-
estimation performance index, is computed as 1.232 × 10−5 . quence, Pka,b,c and Qa,b,c
k are null, and the columns of the
The estimation results are shown in Table I. jacobian will be formed by entries of Ybus (Gkm and Bkm ),
One can notice that voltage magnitudes are correctly es- cos(±120◦ ) = −0.5, cos(0◦ ) = 1, sin(−120◦ ) = −0.866,
timated, but phase angles on the other hand present small sin(120◦ ) = 0.866, and sin(0◦ ) = 0.
differences. It is important to mention that true phase angles Analyzing the active power injection equations (15), (16),
were shifted equally in order to make bus 1 phase a be zero and (17), the entries in the jacobian will be a combination of
degree. The differences are expected since the true bus voltage the terms in (21) and (22) and will the have the form shown
at bus 1 is not perfectly balanced, yet it is assumed so in order in (23):
to use it as the reference bus.
Γkm = 0.866.Gkm + 0.5.Bkm (21)
B. Approach II - Using angle of one phase as the reference
One way to avoid the discrepancies caused by the above Υkm = −0.866.Gkm + 0.5.Bkm (22)
approach is to select only one phase angle as a reference,
allowing the other two phases to converge to the angles
dictated by the measurements. This should allow estimation of 
Baa Γab Υac −Baa −Γab −Υac

the phase imbalances at the reference bus. This is the second
−Υab −Bbb −Γbc 
 
approach, and the difference here compared to the first one  Υab Bbb Γbc
 
is that only one column is removed from the measurement ∂Pk  Γ
 ac Υbc Bcc −Γac −Υbc −Bcc 
=

jacobian before forming the gain matrix. This is the same as ∂θ −Baa −Γab −Υac Baa Γab Υac 

done in positive sequence estimators. However, the problem  
 −Υab −Bbb −Γbc Υab Bbb Γbc 
 
with this approach is that it yields a singular jacobian matrix
evaluated at the flat start. Thus, state estimation problem −Γac −Υbc −Bcc Γac Υbc Bcc
becomes not solvable. (23)
The reason for having a non-solvable system is due to It is evident from (23) that the first three columns are
the active part of measurement jacobian, i.e. some of the the same as the last ones with the opposite sign. Therefore,
columns corresponding to the voltage phase angles are linearly eliminating only one column is not enough to render a non-
dependent. The columns associated with voltage magnitudes singular jacobian matrix. Thus, two more columns must be
however are linearly independent due to the voltage magnitude removed to make it full rank, but this implies knowing the
measurements placed at each phase. The equations describing other two phase angles in advance which is in general not
the entries of the measurement jacobian are shown below to possible. Note that, the reactive power injections have similar
illustrate the creation of linearly dependent columns at flat forms and they will yield a jacobian with the same structure
start. with three linearly dependent columns.
IEEE TRANSACTIONS ON POWER SYSTEMS 6

TABLE II: 2-Bus system states - Approach II

Busphase True States Estimated States

1a 0.9972 0.0000 0.9972 0.0000

b
1 0.9985 -119.8828 0.9985 -119.8828
c
1 0.9994 120.1602 0.9994 120.1602
Fig. 2: 2-Bus system with Virtual Bus
a
2 0.9934 -0.1927 0.9934 -0.1927
2b 0.9995 -119.9997 0.9995 -119.9997 TABLE III: 2-Bus system states - Approach III
c
2 0.9988 120.2226 0.9998 120.2226
Busphase True States Estimated States

1a 0.9972 -0.2205 0.9972 -0.2205

b
1 0.9985 -120.1033 0.9985 -120.1033
2) Flow Equations: The derivatives of active power flow c
measurements with respect to θ are shown below. Reactive 1 0.9994 119.9397 0.9994 119.9397
a
power equations have similar structure, and thus not shown to 2 0.9934 -0.4132 0.9934 -0.4132
save space. 2 b
0.9995 -120.2201 0.9995 -120.2201
2c 0.9988 120.0021 0.9998 120.0021
ph
∂Pkm X
= Vkph ph,p
Vmp (Bkm ph,p
cosθkm − Gph,p ph,p
km sinθkm )
∂θkph p∈Ω
X C. Approach III - Using a three-phase virtual bus reference
ph,p
−Vkph Vkp (B 0km cosθkk
ph,p
− Gph,p ph,p
km sinθkk ) The takeaway from the discussion of approaches I and II
p∈N is that a three phase reference is necessary to render a non-
(24) singular jacobian matrix and find an accurate state estimation
ph
∂Pkm 0ph,p
solution. However, in general a balanced three phase reference
= −Vkph Vkp (Gph,p ph,p ph,p
km sinθkk − B km cosθkk )
(25)
∂θkp bus may not exist, particularly for distribution networks which
ph are very unbalanced. Thus, a third and new approach is needed
∂Pkm
ph
= Vkph Vmph (Gph,ph ph,ph
km sinθkm
ph,ph
− Bkm ph,ph
cosθkm ) and will be outlined next.
∂θm In approach III, a balanced three phase reference bus is
(26) proposed to be used, but this bus is created as a virtual bus
ph
∂Pkm behind the substation (feeder head) bus. This is actually not
p = Vkph Vmp (Gph,p ph,p ph,p ph,p
km sinθkm − Bkm cosθkm )
(27)
∂θm the only choice, but a convenient one. In general, one can
where N stands for the set of adjacent phases, excluding phase choose any bus in a distribution system where there is an
ph; e.g. if ph = a then N is the set of phases b and c. energy source and place a virtual bus behind it.
At the newly created virtual bus, all phase angles are fixed
For overhead distribution lines, the line charging suscep- and displaced 120◦ apart and voltage magnitudes are assumed
tance is very small and can be neglected. In this case, at the to be unknown. This follows the same logic used in the
flat start (24) will be identical to (26), with the opposite sign, modeling of synchronous generators in three-phase power flow
as well as (25) and (27). As a consequence the columns will earlier [47] [50].
be linearly dependent as in the case of injections. It is evident The virtual bus is connected to the substation bus via a
from the above discussion that in general a full rank jacobian three phase balanced impedance. This impedance might be the
matrix will not be obtained using a single phase reference for Thevenin equivalent from the sub-transmission network at the
a three-phase state estimator. point of connection, it can be determined by using short-circuit
The same problem is reported in [49]. Hence, the authors values, or using a known synchronous generator parameter in
suggest to use a different strategy by using a three-phase case the feeder is isolated from the grid and operating as a
reference at the flat start. After that, the it is switched back to microgrid. The 2-bus system with the virtual bus is shown in
single phase reference. Fig. 2.
The original power injection measurements at bus 1 are
Using this strategy, the estimator converges in 4 iterations converted into power flow measurements on the virtual branch,
with a J-index of 1.772 × 10−19 . The estimation results are and substituted by null injection measurements, which are
shown in Table II. treated as equality constraints. The state estimator yields a
Approach II successfully estimates the phase angles when solution after 3 iterations with a J-index of 2.846 × 10−13 .
compared against the true solution. However, such approach The estimation results are shown at Table III, where it is clear
presents problems when applied to larger systems and when that phase angles at bus 1 are perfectly estimated.
measurements are not perfect, as it will be shown in the results For approach III, the phase angles of true states were not
section. equally shifted to make bus 1 phase a be zero degree. These
IEEE TRANSACTIONS ON POWER SYSTEMS 7

are the phase angles obtained from the power flow solution, TABLE IV: 13-node feeder - Mean Squared Errors Normal
and approach III is able to correctly estimate the states with Operation
a three-phase reference. App. V θ Combined
I 6.9305 × 10−9 4.6965 × 10−7 2.3829 × 10−7
VI. S IMULATION R ESULTS II 6.8845 × 10−9 4.8048 × 10−6 2.4058 × 10−6
This section shows the simulations results of executing III 6.9278 × 10−9 4.7873 × 10−7 2.4283 × 10−7
the developed three-phase state estimator on two IEEE test
TABLE V: 13-node feeder - Phase Angles Normal Operation
feeders: (i) IEEE 13-node feeder; (ii) IEEE 123-node feeder;
(iii) IEEE 8500-node feeder. The simulations were performed True States App. I App. II App. III
on a personal computer with an Intel R CoreTM i7-7820HQ −0.0084◦ −0.0084◦ −0.0084◦ −0.0085◦
2.90 GHz processor and 16 GBytes of memory. −120.0073◦ −120.0084◦ −120.0930◦ −120.0073◦
These feeders are already modeled in OpenDSS (can be 119.9909◦ 119.9916◦ 119.8409◦ 119.9910◦
found in [51]) and three-phase power flow is executed to
generate true states and measurements. For each of them, two
scenarios are tested. One using original models as found in It is important to mention that for approach II the measure-
[51] with the source modeled with small impedance values ment jacobian matrix is singular at the flat start. Thus, the flat
simulating an infinite bus. A second scenario where the start strategy is applied to get a converged solution. The phase
source impedance is intentionally made higher to force a more angles of nodes at the root bus are given in Table V.
unbalanced solution. This is equivalent of simulating these The phase angles for approaches I and II are adjusted
feeders being fed by a generator instead of connected to the using the true solution of phase a for easy comparison. For
sub transmission grid, operating like an isolated system or a approach II, unlike the case with 2-bus system with perfect
microgrid. measurements where the phase angle estimates at the root bus
For all scenarios, a set of random Gaussian noise with a were very accurate, here it is observed that the angles deviate
magnitude of 1 × 10−4 is generated and added to the perfect from the true solution. Whereas, approach III provides very
measurements to create a more realistic case. Furthermore, the accurate estimates with small errors, in the order of 1 × 10−4
same set of measurement noise is applied for all scenarios for degrees.
a fair comparison. The number of iterations and time to converge are shown
It is assumed that only the substation has telemetered in Table VI. It is observed that approach III converges with
measurements, which means that the only power flow and less iterations and yielding slightly better J-index.
voltage magnitude measurements available are at the first When a higher source impedance is used to simulate a
branch and the feeder head (root bus). To ensure the system microgrid operation, the voltage imbalance increases in the
is observable, forecasted measurements are added at the load root bus and the differences among the three approaches
buses, and null injections are included at the buses without are more visible. The MSE values and voltage angles for
loads. The values of standard deviations used for telemetered microgrid simulation are given in Tables VII and VIII. Table
and forecasted are 1 × 10−3 and 1 × 10−2 , respectively. The IX gives the number of iterations to converge, J-index, and
convergence tolerance is 1 × 10−5 and a limit of 20 iterations processing for microgrid operation.
is used. The metrics used for comparison purposes are: From these simulation results it is evident that the proposed
• Mean Squared Error (MSE) of the estimated states approach is the only one capable of providing accurate esti-
• Voltage angles at the root bus mates at the root bus. Approach II fails to estimate the correct
• Number of iterations to converge phase angles, although the angles at phases b and c are freely
• Sum of the weighted residuals (J-index) estimated.
• Processing time

TABLE VI: 13-node feeder - Iteration, J-index, and Processing

A. IEEE 13-node feeder Time Normal Operation

This circuit model is very small and operates at 4.16 kV. App. No Iterations J-index Processing Time (s)
It is used to test common features of distribution analysis I 4 0.2245 0.1329
software, being characterized by short lines and relatively high II 4 0.2257 0.1392
loading. There is a single voltage regulator at the substation, III 3 0.1972 0.1149
and it is composed of overhead and underground lines, shunt
capacitors, an in-line transformer, and unbalanced loads [45]. TABLE VII: 13-node feeder - Mean Squared Errors Microgrid
The MSE values from the three approaches are given in Operation
Table IV. In this case, the source is modeled with a tiny
App. V θ Combined
impedance mimicking the normal operation connected to the
grid. The MSE are computed in three ways, considering only I 6.9390 × 10−9 2.0814 × 10−6 1.0442 × 10−6
voltage magnitude estimates, only voltage angles, and both II 6.9069 × 10−9 5.0232 × 10−6 2.5150 × 10−6
combined. III 6.9511 × 10−9 5.0171 × 10−7 2.5433 × 10−7
IEEE TRANSACTIONS ON POWER SYSTEMS 8

TABLE VIII: 13-node feeder - Phase Angles Microgrid Op- TABLE XII: 123-node feeder - Iteration, J-index, and Process-
eration ing Time Normal Operation
True States App. I App. II App. III App. No Iterations J-index Processing Time (s)
−0.8286◦ −0.8286◦ −0.8286◦ −0.8366◦ I 4 0.5259 5.4191
−120.7352◦ −120.8286◦ −120.8227◦ −120.7349◦ II n/a n/a n/a
119.1063◦ 119.1714◦ 118.9529◦ 119.1168◦ III 3 0.4992 4.0461

TABLE IX: 13-node feeder - Iteration, J-index, and Processing TABLE XIII: 123-node feeder - Mean Squared Errors Micro-
Time Microgrid Operation grid Operation

App. No Iterations J-index Processing Time (s) App. V θ Combined

I 4 0.2249 0.1379 I 8.0225 × 10−8 1.8248 × 10−5 9.1641 × 10−6
II 4 0.2258 0.1346 II n/a n/a n/a
III 4 0.1931 0.1531 III 7.9366 × 10−8 4.1020 × 10−6 2.0907 × 10−6

B. IEEE 123-node feeder C. IEEE 8500-node feeder

The 8500-node feeder is a very large scale system derived
The IEEE 123 node test feeder is system with a decent
from a real distribution circuit in the US. The Test Feeders
size for simulation purposes. It has overhead and underground
Working Group (WG) of the Distribution System Analysis
lines, unbalanced loading with constant current, impedance,
Subcommittee of the Power Systems Analysis, Computing
and power, four voltage regulators, shunt capacitor banks,
and Economics (PSACE) Committee has changed some of
and multiple switches. This is a circuit characterized as
its parameters to make the case more interesting, providing
“well-behaved” with minimal convergence problems [45]. The
a good benchmark for research purposes [45]. This feeder in-
results of first scenario, the normal operation, are shown in
cludes many elements found in distribution networks in North
Tables X, XI, and XII.
America such as multiple regulators, single-phase capacitors,
Despite the use of the flat start strategy, approach II does not feeder secondaries, and service transformers. The results of
converge in 20 iterations. Hence, its results are not available. first scenario, the normal operation, are shown in Tables XVI,
Approaches I and III took 4 and 3 iterations to converge XVII, and XVIII.
presenting similar results in terms of MSE values and J-index.
Although approach II converges, the solution is completely
Approach III is capable of estimating the phase angles at the
different than the true states. However, its J-index is small,
root bus with good accuracy.
showing that estimated states satisfy the measurement set.
Although approaches I and III present similar results in the Approaches I and III present similar results in terms of J-index
normal operation, the latter one starts to show its advantage and MSE values.
when the source bus is made more unbalanced in the microgrid During the normal operation, the voltage at the root bus
operation. Approach III is capable of correctly estimating the is very balanced due the strong connection with the sub-
phase imbalances, as can be seen in Tables XIII, XIV, and transmission network. Thus, approaches I and III present sim-
XV. ilar results. However, when the connection is weaker during
The MSE values and J-index are small for both approaches, the microgrid simulation, the differences are more visible, as
slightly better for the proposed one. As the voltage imbalance shown in Tables XIX, XX, and XXI.
at the root bus increases, only approach III is capable of Approach II once again converges in 14 iterations, with a
providing better estimates.

TABLE XIV: 123-node feeder - Phase Angles Microgrid

TABLE X: 123-node feeder - Mean Squared Errors Normal Operation
Operation
True States App. I App. II App. III
App. V θ Combined −1.0133◦ −1.0133◦ n/a −1.0253◦
I 7.9350 × 10−8 3.9956 × 10−6 2.0375 × 10−6 −120.7184◦ −120.0133◦ n/a −120.7205◦
II n/a n/a n/a 119.1237◦ 118.9867◦ n/a 119.1276◦
III 7.9048 × 10−8 4.0297 × 10−6 2.0549 × 10−6
TABLE XV: 123-node feeder - Iteration, J-index, and Process-
TABLE XI: 123-node feeder - Phase Angles Normal Operation ing Time Microgrid Operation
True States App. I App. II App. III App. No Iterations J-index Processing Time (s)
−0.0102◦ −0.0102◦ n/a −0.0101◦ I 4 0.5201 5.2458
−120.0072◦ −120.0102◦ n/a −120.0072◦ II n/a n/a n/a
119.9912◦ 119.9898◦ n/a 119.9912◦ III 4 0.4938 5.3293
IEEE TRANSACTIONS ON POWER SYSTEMS 9

TABLE XVI: 8500-node feeder - Mean Squared Errors Nor- TABLE XX: 8500-node feeder - Phase Angles Microgrid
mal Operation Operation
App. V θ Combined True States App. I App. II App. III
I 5.3398 × 10−6 2.9488 × 10−4 1.5011 × 10−4 −2.8597◦ −2.8597◦ −2.8597◦ −2.8626◦
II 5.3335 × 10−4 4.1179 2.0592 −122.7995◦ −122.8597◦ 15.7623◦ −122.8008◦
III 5.3218 × 10−6 2.9153 × 10−4 1.48436 × 10−4 117.1093◦ 117.1403◦ 292.0141◦ 117.1023◦

TABLE XVII: 8500-node feeder - Phase Angles Normal TABLE XXI: 8500-node feeder - Iteration, J-index, and Pro-
Operation cessing Time Microgrid Operation
True States App. I App. II App. III App. No Iterations J-index Processing Time (s)
−0.0288◦ −0.0288◦ −0.0288◦ −0.0288◦ I 9 0.0644 26.1323
−120.0282◦ −120.0288◦ 18.0591◦ −120.0282◦ II 14 0.0609 39.8192
119.9717◦ 119.9712◦ 296.1279◦ 119.9717◦ III 9 0.0507 29.4274

small J-index, but huge MSE values. Phase angle estimates at ACKNOWLEDGMENTS
the root bus are also completely different than the true solution.
Approaches I and III present similar results in terms of J-index This material is based upon work supported by the U.S.
and MSE values, where the proposed approach has a slightly Department of Energy’s Office of Energy Efficiency and
better outcome. The microgrid operation evidences how the Renewable Energy (EERE) under Solar Energy Technologies
proposed approach excels compared to conventional ones. In Office (SETO) Agreement Number EE0008005. Engineering
this case the voltage imbalance increases, and only the use of Research Center shared facilities supported by the Engineering
virtual bus allows to accurately estimate it. Research Center Program of the National Science Foundation
and the Department of Energy under NSF Award Number
VII. C ONCLUDING R EMARKS EEC- 1041877 and the CURENT Industry Partnership Pro-
This paper identifies a shortcoming of the existing distribu- gram.
tion state estimation algorithms where one of the system buses
is implicitly assumed to have perfectly balanced three-phase R EFERENCES
voltages. While this may be the case for typical distribution
substations which have balanced voltages imposed by the [1] M.-S. Chen and W. E. Dillon, “Power system modeling,” Proceedings
of the IEEE, vol. 62, no. 7, pp. 901–915, 1974.
strongly balanced transmission networks behind them, the [2] F. C. Schweppe and J. Wildes, “Power system static-state estimation, part
same may not be always true when solving active distribution i: Exact model,” IEEE Transactions on Power Apparatus and systems,
networks with distributed energy sources and unbalanced no. 1, pp. 120–125, 1970.
[3] F. C. Schweppe and D. B. Rom, “Power system static-state estimation,
loads. The paper proposes a simple yet effective way to part ii: Approximate model,” IEEE Transactions on Power Apparatus
overcome this limitation and allows the solution of all the and Systems, no. 1, pp. 125–130, 1970.
mixed-phase unbalanced distribution system buses including [4] F. C. Schweppe, “Power system static-state estimation, part iii: Imple-
mentation,” IEEE Transactions on Power Apparatus and systems, no. 1,
the feeder head to have unbalanced voltage solutions. The pp. 130–135, 1970.
method is illustrated first by tutorial examples and then [5] I. Roytelman and S. Shahidehpour, “State estimation for electric power
validated on small, medium and very large scale distribution distribution systems in quasi real-time conditions,” IEEE Transactions
on Power Delivery, vol. 8, no. 4, pp. 2009–2015, 1993.
systems containing untransposed three phase, as well as two [6] N. Vempati, R. Shoults, M. Chen, and L. Schwobel, “Simplified feeder
phase and single phase feeder sections and unbalanced loads. modeling for loadflow calculations,” IEEE transactions on power sys-
tems, vol. 2, no. 1, pp. 168–174, 1987.
[7] W. H. Kersting, Distribution system modeling and analysis. CRC press,
TABLE XVIII: 8500-node feeder - Iteration, J-index, and 2012.
Processing Time Normal Operation [8] T.-H. Chen, M.-S. Chen, T. Inoue, P. Kotas, and E. A. Chebli, “Three-
phase cogenerator and transformer models for distribution system analy-
App. No Iterations J-index Processing Time (s) sis,” IEEE Transactions on Power Delivery, vol. 6, no. 4, pp. 1671–1681,
1991.
I 9 0.0699 26.3447 [9] T.-H. Chen and J.-D. Chang, “Open wye-open delta and open delta-open
II 14 0.0655 38.1251 delta transformer models for rigorous distribution system analysis,” in
IEE Proceedings C (Generation, Transmission and Distribution), vol.
III 10 0.0552 29.0648 139, no. 3. IET, 1992, pp. 227–234.
[10] M. Gorman and J. Grainger, “Transformer modelling for distribution
TABLE XIX: 8500-node feeder - Mean Squared Errors Mi- system studies. ii. addition of models to y/sub bus/and z/sub bus,” IEEE
crogrid Operation transactions on power delivery, vol. 7, no. 2, pp. 575–580, 1992.
[11] R. Dugan and S. Santoso, “An example of 3-phase transformer modeling
App. V θ Combined for distribution system analysis,” in Transmission and Distribution
Conference and Exposition, 2003 IEEE PES, vol. 3. IEEE, 2003, pp.
I 4.0263 × 10−6 1.3999 × 10−4 7.2012 × 10−5 1028–1032.
II 5.7649 × 10−4 4.0807 2.0406 [12] H. Kim and A. Abur, “State estimation for three phase power networks,”
in Proceedings of the 26th Annual North American Power Symposium,
III 4.1198 × 10−6 1.5302 × 10−4 7.8572 × 10−5 1994, pp. 27–28.
IEEE TRANSACTIONS ON POWER SYSTEMS 10

[13] M. E. Baran and A. W. Kelley, “State estimation for real-time monitoring [36] M. Pau, P. A. Pegoraro, and S. Sulis, “Efficient branch-current-based dis-
of distribution systems,” IEEE Transactions on Power Systems, vol. 9, tribution system state estimation including synchronized measurements,”
no. 3, pp. 1601–1609, 1994. IEEE Transactions on Instrumentation and Measurement, vol. 62, no. 9,
[14] C. Hansen and A. Debs, “Power system state estimation using three- pp. 2419–2429, 2013.
phase models,” IEEE transactions on power systems, vol. 10, no. 2, pp. [37] U. Kuhar, M. Pantoš, G. Kosec, and A. Švigelj, “The impact of model
818–824, 1995. and measurement uncertainties on a state estimation in three-phase
[15] C. Lu, J. Teng, and W.-H. Liu, “Distribution system state estimation,” distribution networks,” IEEE Transactions on Smart Grid, vol. 10, no. 3,
IEEE Transactions on Power systems, vol. 10, no. 1, pp. 229–240, 1995. pp. 3301–3310, 2018.
[16] K. Li, “State estimation for power distribution system and measurement [38] E. Nogueira, R. Portelinha, E. Lourençço, O. Tortelli, and B. Pal,
impacts,” IEEE Transactions on Power Systems, vol. 11, no. 2, pp. 911– “A novel approach to power system state estimation for transmission
916, 1996. and distribution systems,” IET Generation Transmission & Distribution,
[17] W.-M. Lin and J.-H. Teng, “State estimation for distribution systems with vol. 13, pp. 1970–1978, 2019.
zero-injection constraints,” in Proceedings of Power Industry Computer [39] A. Primadianto and C.-N. Lu, “A review on distribution system state
Applications Conference. IEEE, 1995, pp. 523–529. estimation,” IEEE Transactions on Power Systems, vol. 32, no. 5, pp.
3875–3883, 2016.
[18] N. Singh, E. Kliokys, H. Feldmann, R. Kussel, R. Chrustowski, and
[40] F. Ahmad, A. Rasool, E. Ozsoy, R. Sekar, A. Sabanovic, and M. Elitaş,
C. Joborowicz, “Power system modelling and analysis in a mixed energy
“Distribution system state estimation-a step towards smart grid,” Renew-
management and distribution management system,” IEEE transactions
able and Sustainable Energy Reviews, vol. 81, pp. 2659–2671, 2018.
on power systems, vol. 13, no. 3, pp. 1143–1149, 1998.
[41] G. N. Korres, N. D. Hatziargyriou, and P. J. Katsikas, “State estimation
[19] S. Zhong and A. Abur, “Effects of nontransposed lines and unbalanced
in multi-microgrids,” European transactions on electrical power, vol. 21,
loads on state estimation,” in Power Engineering Society Winter Meeting,
no. 2, pp. 1178–1199, 2011.
2002. IEEE, vol. 2. IEEE, 2002, pp. 975–979.
[42] D. Issicaba, A. S. Costa, and J. L. Colombo, “Real-time monitoring
[20] A. S. Meliopoulos and F. Zhang, “Multiphase power flow and state of points of common coupling in distribution systems through state
estimation for power distribution systems,” IEEE Transactions on Power estimation and geometric tests,” IEEE Transactions on Smart Grid,
Systems, vol. 11, no. 2, pp. 939–946, 1996. vol. 7, no. 1, pp. 9–18, 2015.
[21] M. E. Baran and A. W. Kelley, “A branch-current-based state estimation [43] P. M. De Oliveira-De Jesus and C. H. Antunes, “A detailed network
method for distribution systems,” IEEE transactions on power systems, model for distribution systems with high penetration of renewable
vol. 10, no. 1, pp. 483–491, 1995. generation sources,” Electric Power Systems Research, vol. 161, pp.
[22] W.-M. Lin, J.-H. Teng, and S.-J. Chen, “A highly efficient algorithm in 152–166, 2018.
treating current measurements for the branch-current-based distribution [44] A. Gjelsvik, S. Aam, and L. Holten, “Hachtel’s augmented matrix
state estimation,” IEEE Transactions on Power Delivery, vol. 16, no. 3, method-a rapid method improving numerical stability in power system
pp. 433–439, 2001. static state estimation,” IEEE Transactions on Power Apparatus and
[23] H. Wang and N. N. Schulz, “A revised branch current-based distribution Systems, no. 11, pp. 2987–2993, 1985.
system state estimation algorithm and meter placement impact,” IEEE [45] IEEE PES DSA Working Group, IEEE PES AMPS DSAS Test Feeder
Transactions on Power Systems, vol. 19, no. 1, pp. 207–213, 2004. Working Group, (accessed August 27, 2018). [Online]. Available:
[24] M. Pau, F. Ponci, A. Monti, S. Sulis, C. Muscas, and P. A. Pegoraro, “An http://sites.ieee.org/pes-testfeeders/resources/
efficient and accurate solution for distribution system state estimation [46] K. A. Birt, J. J. Graffy, J. D. McDonald, and A. H. El-Abiad, “Three
with multiarea architecture,” IEEE Transactions on Instrumentation and phase load flow program,” IEEE Transactions on Power Apparatus and
Measurement, vol. 66, no. 5, pp. 910–919, 2017. Systems, vol. 95, no. 1, pp. 59–65, 1976.
[25] I. Džafić and R. A. Jabr, “Real time multiphase state estimation in [47] A. Abur, H. Singh, H. Liu, and W. Klingensmith, “Three phase power
weakly meshed distribution networks with distributed generation,” IEEE flow for distribution systems with dispersed generation,” 14th PSCC,
Transactions on Power Systems, vol. 32, no. 6, pp. 4560–4569, 2017. Sevilla, vol. 11, no. 3, 2002.
[26] R. Singh, B. C. Pal, and R. A. Jabr, “Statistical representation of distri- [48] R. C. Dugan and T. E. McDermott, “An open source platform for
bution system loads using gaussian mixture model,” IEEE Transactions collaborating on smart grid research,” in Power and Energy Society
on Power Systems, vol. 25, no. 1, pp. 29–37, 2009. General Meeting, 2011 IEEE. IEEE, 2011, pp. 1–7.
[27] M. E. Baran, J. Zhu, and A. W. Kelley, “Meter placement for real- [49] M. C. de Almeida, E. N. Asada, and A. V. Garcia, “Effects of load
time monitoring of distribution feeders,” IEEE Transactions on Power imbalance and system asymmetry on three-phase state estimation,” in
systems, vol. 11, no. 1, pp. 332–337, 1996. 2006 IEEE Power Engineering Society General Meeting. IEEE, 2006,
[28] R. Singh, B. C. Pal, R. A. Jabr, and R. B. Vinter, “Meter placement for pp. 6–pp.
distribution system state estimation: An ordinal optimization approach,” [50] J. Arrillaga, C. Arnold, and B. Harker, Computer modelling of electrical
IEEE Transactions on Power Systems, vol. 26, no. 4, pp. 2328–2335, power systems. Wiley Online Library, 2001, vol. 2.
2011. [51] EPRI Distribution System Simulator , OpenDSS Source-
Forge, (accessed November 11, 2019). [Online]. Available:
[29] A. Shafiu, N. Jenkins, and G. Strbac, “Measurement location for state
https://sourceforge.net/projects/electricdss/
estimation of distribution networks with generation,” IEE Proceedings-
Generation, Transmission and Distribution, vol. 152, no. 2, pp. 240–246,
2005.
[30] A. Gómez-Expósito, C. Gómez-Quiles, and I. Džafić, “State estimation
in two time scales for smart distribution systems,” IEEE Transactions
on Smart Grid, vol. 6, no. 1, pp. 421–430, 2014. Andre L. Langner (S’18) received the B.S. and M.Sc. degrees from the
Federal University of Parana (UFPR), Curitiba, Brazil, in 2014 and 2016,
[31] A. Alimardani, F. Therrien, D. Atanackovic, J. Jatskevich, and E. Vaa-
respectively. He is with Siemens Energy in Minneapolis/MN, USA, working
hedi, “Distribution system state estimation based on nonsynchronized
on transmission network applications, and also with Northeastern University
smart meters,” IEEE Transactions on Smart Grid, vol. 6, no. 6, pp.
in Boston/MA, USA, working towards the Ph.D. degree in Electrical Engi-
2919–2928, 2015.
neering.
[32] S.-C. Huang, C.-N. Lu, and Y.-L. Lo, “Evaluation of ami and scada data
synergy for distribution feeder modeling,” IEEE Transactions on smart
Grid, vol. 6, no. 4, pp. 1639–1647, 2015.
[33] M. Huang, Z. Wei, G. Sun, and H. Zang, “Hybrid state estimation for
distribution systems with ami and scada measurements,” IEEE Access, Ali Abur (F’03) received the B.S. degree in electrical engineering from the
vol. 7, pp. 120 350–120 359, 2019. Orta Dog̃u Teknik Üniversitesi, Ankara, Turkey and the M.S. and Ph.D.
[34] R. S. Silva, F. M. Laburu, and M. C. de Almeida, “On the use of µpmu degrees from The Ohio State University, Columbus, OH, USA. He is cur-
for state estimation in distribution systems,” in 2017 IEEE Power & rently a Professor with the Electrical and Computer Engineering Department,
Energy Society General Meeting. IEEE, 2017, pp. 1–5. Northeastern University, Boston, MA, USA
[35] J. Liu, J. Tang, F. Ponci, A. Monti, C. Muscas, and P. A. Pegoraro,
“Trade-offs in pmu deployment for state estimation in active distribution
grids,” IEEE transactions on Smart Grid, vol. 3, no. 2, pp. 915–924,
2012.