1

Tracking State Estimation in Distribution Networks

Using Distribution-level Synchrophasor Data
Mohammad Farajollahi, Alireza Shahsavari, and Hamed Mohsenian-Rad
Department of Electrical and Computer Engineering, University of California, Riverside, CA, USA
e-mails: {mfara006, ashah023}@ucr.edu, and hamed@ece.ucr.edu

Abstract—State estimation is often a challenging task in dis- usage of TSE in distribution system is developed to incorpo-
tribution systems due to numerous nodes and limited number rate unsynchronized measurements within DSSE. The synergy
of measurement sources. As a result, distribution system state of smart meter and supervisory control and data acquisition
estimation is often not real-time; instead, it is done once every 5
to 15 minutes, e.g., when smart meter measurements are available. (SCADA) data for DSSE is proposed by [3], where the archived
Accordingly, this paper proposes a novel method for tracking state customer meter data and the SCADA data with the same time
estimation in distribution systems to update the system states stamp are used off-line to estimate the states at feeder buses.
following an event, e.g., a sudden load change, that occurs in The results are then used in the on-line quasi dynamic state
between the 5 to 15 minutes intervals of a typical distribution estimations. In [4], the authors investigate the effect of smart
system state estimation cycle, without the need to rerun the whole
state estimation. This method is of great interest for real-time meter data in real-time state estimation. To do so, they treat
monitoring and control applications. The proposed method uses smart meter data as pseudo-measurement to incorporate them
measurements from as few as only two distribution-level phasor into the real time state estimation, and consider the effect of
measurement units, a.k.a., micro-PMUs, that are installed at the delay in the collected smart meter data on real time DSSE. In
substation and at the end of the main feeder or laterals. The addition, the incorporation of different real-time metering and
method is developed based on the compensation theorem in circuit
theory to generate an equivalent circuit according to the pre-event measurement instruments was scrutinized in [5], [6].
and post-event feeder data in order to update the state estimation Recently, various applications have been reported for micro-
results. The effectiveness of the proposed method is demonstrated PMUs in distribution systems, e.g., see [1], [7]–[9]. Con-
through simulating the IEEE 33-bus test system. sidering the applications that are relevant to DSSE, in [10],
Keywords: Tracking state estimation, micro-PMUs, compensation the authors proposed an efficient branch-current-based DSSE,
theorem, data-driven, distribution system events. which integrates synchronized phasor measurements provided
by PMUs. Haughton et al [11] developed a linearized, three-
phase, distribution class state estimation algorithm for real-time
I. I NTRODUCTION applications in smart distribution systems. The practical use
Real-time monitoring and control play a crucial role in power of micro-PMUs in standard DSSE algorithms is reported in
distribution networks to achieve optimal and reliable operation. [12]. The integration of micro-PMU data, together with remote
Distribution System State Estimation (DSSE) is a cornerstone terminal unit data and smart meter data, into the DSSE is
tool in advance distribution system operation. It uses the discussed in [13]. To the best of our knowledge, no prior study
measurements, e.g., smart meters or asset sensors, to best has addressed the application of micro-PMU data in TSE.
approximate the states of the system which best fit the available In this paper, we propose a novel TSE method to be added
measurements. As the system scales up, the computational to the standard DSSE, so as to make use of voltage and
burden of DSSE grows to the extent that the computational time current synchrophasor data in order to update the system states
may exceed the rate at which new measurements are obtained. following a change in a system element. The changes in the
This is particularly challenging when a subset of measure- system can be detected by the voltage and current phasors
ments come from the emerging advanced sensor devices, such recorded through micro-PMUs. The essence of the proposed
as distribution-level phasor measurement units, a.k.a, micro- method is based on the analysis of the equivalent-circuit for
PMUs [1]. However, the states of the system do change on a the distribution feeder, by making use of compensation theo-
continuous basis; thus, updating the DSSE results is necessary rem from circuit theory [14], according to the pre-event and
to enhance situational awareness at distribution level, as well post-event micro-PMUs measurement. Our approach is highly
as to carry out more efficient control and operation. practical because it requires using as few as only two micro-
A technique called Tracking State Estimation (TSE) can PMUs. The micro-PMUs are proposed to be installed at the
address the above issues. TSE does not require running a full substation and at the end of the main feeder or laterals.
state estimation. It rather only fine tunes the states of the system
at occasions during the standard state estimation cycles. II. P ROPOSED M ETHOD
The concept of TSE was first introduced by F. C. Schweppe
In this paper, we seek to develop a TSE method to update
in 1980 [2] for transmission system state estimation. The
the results in distribution system state estimation after an event
This work was supported in part by NSF grant 1462530 and DoE Grant occurs. Here, distribution system events are defined as any
0008001. The corresponding author is H. Mohsenian-Rad. change in system parameters which does not interrupt the
normal operation of the system, such as a load changing in the equivalent circuit, the voltages of the buses equipped with
a certain location. Such events are detected by monitoring the micro-PMUs are known, i.e., we have:
voltage and current measurements that are recorded by micro-
PMUs. For example, by applying the method in [7], one can ∆V1 = ∆V u (3)
not only detect the distribution system events but also identify
their root causes, i.e., to indicate whether each event occurred and
on the distribution feeder of interest; or whether it occurred ∆Vn = ∆V d , (4)
in some other parts of the network, e.g., at transmission or
sub transmission systems. Accordingly, for the rest of this where ∆V1 and ∆Vn indicate the difference voltage phasors of
paper, we assume that a method such as the one in [7] is bus 1 and bus n which are home to the two micro-PMUs.
already applied to the micro-PMU measurements to narrow The proposed application of the compensation theorem is to
down our focus in TSE to only those events that have root help estimating the location and injection of the current source
causes in the distribution system and thus they require updating associated with the event to best fit the the changes captured via
the distribution system state estimation results. micro-PMUs. The importance of the compensation theorem is
to allow analyzing an event through examining such equivalent
circuit, which makes the analysis easier and faster.
A. Equivalent Circuit
An event in a circuit, such as a change in an element, can
change all or a subset of nodal voltages and branch currents B. Network Model
along the circuit. According to the compensation theorem in
circuit theorem [14, pp. 177], the amount of such changes in Let ∆V = [∆V1 , · · · , ∆Vn ]T and ∆I = [∆I1 , · · · , ∆In ]T
the nodal voltages and branch currents can be obtained through denote the voltage and current difference vectors. Even with
an equivalent circuit, in which the changed element is replaced no knowledge on the location of the event, we can write the
with a current source that injects current at a level equal to the Kirchhoff’s current law (KCL) for this equivalent circuit as:
amount of change in the current going through the element;
∆I = (Y bus + e1 eT1 y u + en eTn y d )∆V (5)
and all sources are replaced with their internal impedances.
Consider a segment in a power distribution network, such where, Y bus is the standard nodal admittance matrix of the
as the one shown in Fig. 1. Assume that two micro-PMUs equivalent circuit. Let yii0 denote admittance line between two
are installed1 at the upstream and downstream of this segment; nodes i and i0 . In this regard, Y bus is an n × n complex-
thus distinguishing this part of the network from the rest of valued matrix whose off-diagonal elements are Yiibus = −yii0
0
the power system . Suppose n denotes the number of buses bus
P
and diagonal elements are Yii = i0 6=i yii0 . The equivalent
between these two micro-PMUs, enumerated from 1 to n. admittance of the upstream and downstream networks are
Suppose an event occurs at bus k, where k ∈ {1, ...n}. From integrated into the model via e1 eT1 y u and en eTn y d , where ei is
compensation theorem, an equivalent circuit can be constructed an n×1 standard basis vector, i.e., ei = [0, · · · , 0, 1, 0, · · · , 0]T .
for this power system in which a current source is placed
with current ∆Ik at bus k, and the upstream and downstream
networks are replaced with their equivalent impedances. The C. Lease Squares Estimation
equivalent impedance for the upstream network is obtained as:
Assume that the initial states of the system are given through
u ∆I u the most recent cycle of the standard DSSE application, i.e.,
y , , (1)
∆V u sometime over the past 15 minutes. Therefore, the loading
where ∆V u and ∆I u indicate the changes in voltage and situation at each node in the system is known prior to the
current before and after the event that are captured by the occurrence of the first event. Without loss of generality, we
micro-PMU installed at the upstream. Similarly, the equivalent assume that all loads are constant-impedance, such that at each
impedance for the downstream network is obtained as: bus i, ∆Vi is equal to Zi ∆Ii , where Zi indicates the impedance
of the load at bus i and is calculated from the most recent state
∆I d of the system. Other types of loads, namely constant-current
yd , , (2)
∆V d and constant-power loads, can also be formulated and similarly
where ∆V d and ∆I d indicate the changes in voltage and cur- integrated into the model, e.g., see [15].
rent captured by the micro-PMU installed at the downstream. The product of bus current injection and bus impedance is
The nodal voltages and branch currents of the buses equipped valid at all buses except for at bus k in which the event occurs.
with micro-PMUs in the presence of this current source are Recall from the compensation theorem that at the event bus,
equal to the changes in nodal voltages and branch currents a current source injects ∆Ik into the equivalent circuit; and
obtained from subtracting pre-event and post-event voltage and hence, the product of bus current and bus impedance is no
current phasors recorded by the micro-PMUs. Therefore, in longer a correct indication of the bus current. Therefore, we can
consider the following least squares (LS) problem to estimate
1 The case with several micro-PMUs is explained in Section II-E. vectors ∆V and ∆I that are induced by the event:
 d
y
u y
∆Ik

∆Iu ∆V1 ∆V2 ∆Vk ∆Vn-1 ∆Vn ∆Id

y12 y(n-1)n
Upstream Downstream
d
u
∆V ∆V
∆I1 ∆I2 ∆Ik ∆In-1 ∆In
Z1 Z2 Zk Zn-1 Zn

Fig. 1. Representation of a network based on compensation theorem equivalent circuit. The targeted distribution network is restricted via several micro-PMUs.

n
where k ? is determined based on the binary variable associated
Minimize
X
kbi (∆Vi − Zi ∆Ii )k
2 with the event bus. Specifically, k ? = i for bus index i for which
∆V,∆I,b
i=1
b? = 0. It is worth to mention that the location of the event,
Subject to Eqs. (3), (4), (5) (6) i.e., k ? , can be determined also through a method proposed in
n
[7], which we use in this paper as a validation for our proposed
method. Algorithm 1 summarized the proposed model for TSE.
X
bi = n − 1,
i=1
Algorithm 1 TSE
T
where b = [b1 , · · · , bn ] indicates the vector of binary vari- 1: Get initial system states.
ables, in which bk serves to check the hypothesis of the event 2: Obtain network model equations using (5).
occurring at bus k. If we assume that the event occurs at 3: Solve LSE (6).
bus k, then we can conclude that ∆Vi = Zi ∆Ii , for any 4: return Updated system states using (7).
i ∈ {1, · · · , n} − {k}. Therefore, the objective function in 5: return Updated impedance of event bus using (8).
(6) is minimized once the binary variable associated with the
true event bus, i.e., bk , is zero. This is enforced by the binary
summation constraint. In addition, the correct solution should
be in line with the network model and the measurements E. Extension for Multiple Micro-PMUs
which are directly recorded by the micro-PMUs. To do so, the The method that was developed in Sections II-A to II-D are
constraints in (3) and (4) serve to check the validity of the based on the assumption that the distribution network segment
solution with the micro-PMU measurements. Also, constraint of interest is restricted by two micro-PMUs at its upstream and
in (5) is used for check the consistency of the solution with downstream. Next, suppose the network segment is restricted
the equivalent circuit model. Here, we make the practical by several micro-PMUs. Each micro-PMU has a downstream,
assumption that the measurements from the micro-PMUs are with the following equivalent impedance of:
precise. However, a large value of the residual in the optimal
∆I j
objective value of the problem in (6) could be an indication for yj , , ∀j ∈ ΩP M U , (9)
bad data measurement. ∆V j
where ΩP M U represent the set of buses which are equipped
D. Updating System State with micro-PMUs, j is an index. In this regard, the network
model in (5) can be modified as:
Let ∆V ? and ∆I ? ,denote the optimal solution of problem
(6). The system voltages and currents are updated as:
 
X
∆I = Y bus + ej eTj y j  ∆V. (10)
V post = V pre + ∆V ?
(7) j∈ΩP M U
I post = I pre + ∆I ?
The rest of the model would be the same as expressed in (6)
where V pre and I pre indicate the bus voltages and injection for the case of two micro-PMUs. Of course, the constraints
currents prior to the event. It is assumed that the most recent with respect to (3) and (4) will also be expanded to include the
system state is given based on the standard state estimation voltage of the buses equipped with micro-PMUs.
carried out for each cycle and several continuous updating in
the cycle. Therefore, the pre-event system state is known and
III. C ASE S TUDY
the system sate is updated following an event occurrence as
described in (7). Moreover, for the next round of the updating, This section demonstrates the effectiveness of the proposed
the impedance of the bus undergoing the event is modified as: TSE method on the 33 bus distribution test system. The single
line diagram of the under-study feeder is shown in Fig. 2, which
Vkpost
? includes a main feeder from bus 1 to bus 18 as well as three
Zk ? = , (8)
Ikpost
?
laterals respectively connected to bus 2, bus 3, and bus 6. The
 19 20 21 22 26 27 28 29 30 31 32 33

µPMU 3

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Sub.

23 24 25

µPMU 1 µPMU 2

Fig. 2. Under-study feeder.

12 6 TABLE I
V R ESULTS ASSOCIATED WITH C HANGES IN B US 8
V Error
10 5
Bus # V pre ∆V ∆I ∆S
8 6880.4 ∠ -0.06 10.3 ∠ 12.85 2.61 ∠ 156.81 49.7+j21.2
8 4

V Error (%)
V (volt)

6 3 TABLE II
R ESULTS FOR D IFFERENT L OAD C HANGING IN S EVERAL B USES
4 2
Max Error
Bus # Load Switching ∆S
in ∆V (%)
2 1
4 120+j80 1.3 120.5+j79.7
8 20+j10 5.0 20.1+j10.5
0 0
1 5 10 15 20 25 30 33 24 200 2.9 199.9+j4.5
Bus Number 30 j200 7.8 12.4+j190.7

Fig. 3. Estimated buses voltage and related errors with the hypotheses of event
at bus 8.
mentioned that the errors greater than 1% correspond to the
buses hosted by first lateral branch. This error can be decrease
load data and line data corresponding to the under-study test by installing a micro-PMU at the end of the lateral. Also, from
system can be found in [16]. comparing errors of bus 1 to 18 in Fig. 3, we can conclude that
To this case study, MATLAB Power System Simulation by moving far away from the monitored buses, generally the
Package (MATPOWER 6.0) is utilized to simulate the under- error values increase.
study test system [17]. Also, we assume that three micro-PMUs Table II shows the results associated with loads curtailment
are deployed at bus 1, 18, and 33, see Fig. 2. Accordingly, in buses 4, 8, 24, and 30. For an individual event at the buses
voltage difference phasors and current difference phasors cor- on the main feeder, i.e., buses 4 and 8, as well as the third
responding to monitored buses are calculated by conducting lateral, i.e., bus 30, there is only one optimal solution for (6).
pre-event and post-event steady state power flows through However, for the event at bus 24, which is hosted by the second
MATPOWER. Next, the obtained results for different loads lateral, there are four possible optimal solutions for (6), i.e.,
switching will be reported followed by sensitivity analyses bi = 0, ∀i ∈ {3, 23, 24, 25} results the minimum value for
to demonstrate the effectiveness of the proposed TSE method objective function. In other words, it can be noticed that the
against network topology data as well as measurement data event has occurred in one of the buses hosted by second lateral,
quality. but it is not precisely determined. In order to distinguish these
buses, another micro-PMU should be installed at the end of the
A. Results second lateral.
Suppose a 50kW+20kVAR load curtailment on the bus 8,
which is about 30% and 20% reduction of nominal loading. Ta- B. Sensitivity Analysis
ble I shows the results associated with estimated load changing In this section, the effectiveness of proposed TSE method
in bus 8. As it can be seen, the load changing is estimated against network topology data as well as measurement accuracy
as 49.7kW+21.2kVAR, which demonstrates the effectiveness will be studied. To such aim, sensitivity analyses will be
of the proposed TSE method, i.e., 1% error in active power conducted by considering line admittance uncertainty, voltage
curtailment and about 6% error in reactive power curtailment. phasor measurement accuracy, and current phasor accuracy.
In addition, Fig. 3 shows the estimated voltage magnitude In order to model uncertainties, a Monte Carlo algorithm is
changes, due to load curtailment in bus 8, obtained from TSE utilized to generate scenarios.
method in the equivalent circuit of under-study network. Also, Table III shows the results obtained from TSE considering
errors of estimated voltage magnitude changes are reported in 3. lines admittance uncertainty. As it can be seen, for line ad-
As it can be seen from figure, the maximum errors in estimated mittances error with 5% standard deviation, the load switching
∆V corresponds to bus 22, which is about 5%. It is worth to bus is precisely located. However, as expected, by increasing
 TABLE III
M ETHOD E FFECTIVENESS VS . L INE A DMITTANCES U NCERTAINTY range of error in lines admittance and measurement accuracy,
TSE results remain in an acceptable range of error, confirming
Line Admittace Correct Neighbors ∆V Error P Error Q Error
Error SD (%) (%) (%) SD (%) SD (%) SD (%) a reliable estimation with potential application in practice.
5 100 0 15.6 0.1 0.2
10 81 18 28.4 0.2 0.5 R EFERENCES
15 61 33 45.8 0.3 0.8
20 37 36 59.1 0.5 1.1 [1] A. von Meier, E. Stewart, A. McEachern, M. Andersen, and
L. Mehrmanesh, “Precision micro-synchrophasors for distribution sys-
tems: A summary of applications,” IEEE Trans. on Smart Grid, vol. 8,
TABLE IV no. 6, pp. 2926 – 2936, Nov. 2017.
M ETHOD E FFECTIVENESS VS . VOLTAGE M EASUREMENTS U NCERTAINTY [2] F. Schweppe and R. Masiello, “A tracking static state estimator,” IEEE
Trans. on Power Apparatus and Systems, no. 3, pp. 1025–1033, 1971.
Voltage Phasor Correct Neighbors ∆V Error P Error Q Error [3] S.-C. Huang, C.-N. Lu, and Y.-L. Lo, “Evaluation of ami and scada data
Error SD (%) (%) (%) SD (%) SD (%) SD (%) synergy for distribution feeder modeling,” IEEE Trans. on Smart Grid,
5 98 2 5.1 0.1 0.5 vol. 6, no. 4, pp. 1639–1647, 2015.
10 76.6 21.8 10.2 0.2 0.9 [4] T. Xygkis, G. Karlis, I. Siderakis, and G. Korres, “Use of near real-
15 61 30.6 15.3 0.3 1.5 time and delayed smart meter data for distribution system load and state
20 44.2 36.6 20.9 0.5 2.0 estimation,” in Proc. of MedPower, Athens, Greece, Jan. 2014.
25 35.2 34.4 24.8 0.6 2.3 [5] X. Han, S. You, F. Thordarson, D. V. Tackie, S. M. Ostberg, O. M.
Pedersen, H. W. Bindner, and N. C. Nordentoft, “Real-time measurements
and their effects on state estimation of distribution power system,” in Proc.
of ISGT Europe, Lyngby, Denmark, 2013.
the lines admittance error, the wrong event location increase, [6] F. Xiaoming, F. Yang, and W. Peterson, “A practical multi-phase distribu-
but most of the wrong identifications are related to identifying tion state estimation solution incorporating smart meter and sensor data,”
in Proc. of IEEE PES General Meeting, San Diego, CA, USA, Jul. 2012.
the neighboring buses as the load switching location, see [7] M. Farajollahi, A. Shahsavari, and H. Mohsenian-Rad, “Location identifi-
third column of Table III .Moreover, table III shows that the cation of distribution network events using synchrophasor data,” in Proc.
maximum error in estimated ∆V is highly sensitive to the of North American Power Symposium, 2017.
[8] A. Shahsavari, M. Farajollahi, E. Stewart, A. von Meier, L. Alvarez,
lines admittances. It is obvious that once the knowledge about E. Cortez, and H. Mohsenian-Rad, “A data-driven analysis of capacitor
the network is not precise, estimation of the network state bank operation at a distribution feeder using micro-pmu data,” in IEEE
is accompanied with error. Finally, the last two columns of PES ISGT, Washington D.C., Apr. 2017.
[9] M. Farajollahi, A. Shahsavari, and H. Mohsenian-Rad, “Location identi-
table III shows the effectiveness of proposed TSE method in fication of high impedance faults using synchronized harmonic phasors,”
estimating load switching active power and reactive power even in Proc. IEEE PES ISGT, Washington D.C., Apr. 2017.
in presence of lines admittance uncertainties. [10] M. Pau, P. A. Pegoraro, and S. Sulis, “Efficient branch-current-based dis-
tribution system state estimation including synchronized measurements,”
Table IV and V show the sensitivity of the proposed TSE IEEE Trans. on Instrumentation and Measurement, vol. 62, no. 9, pp.
method for measurements accuracy. The standard deviation of 2419–2429, 2013.
errors associated with micro-PMUs include both phasor angle [11] D. A. Haughton and G. T. Heydt, “A linear state estimation formulation
for smart distribution systems,” IEEE Trans. on Power Systems, vol. 28,
and magnitude errors. We can clearly see that the error in no. 2, pp. 1187–1195, 2013.
voltage phasors has greater effect on the event location accuracy [12] C. Muscas, M. Pau, P. A. Pegoraro, S. Sulis, F. Ponci, and A. Monti,
than the error in current phasors. However, the estimated value “Multiarea distribution system state estimation,” IEEE Trans. on Instru-
mentation and Measurement, vol. 64, no. 5, pp. 1140–1148, 2015.
of load changing is highly sensitive to the current measurements [13] W. Shaofang and L. GuangYi, “State estimation for active distribution
error. networks based on multi source hybrid measurements,” in Proc. China
Int. Conference on Electricity Distribution, Xi’an, China, Sep. 2016.
[14] K. S. Kumar, Electric circuits and networks. Pearson Education India,
IV. C ONCLUSIONS 2009.
[15] D. He, T. Habetler, M. J. Mousavi, and N. Kang, “A zip model-based
This paper proposed a novel method for Tracking State feeder load modeling and forecasting method,” in Proc. Power and Energy
Estimation (TSE). The method is built on equivalent circuit Society General Meeting, Vancouver, BC, Canada, Jul. 2013.
[16] M. E. Baran and F. F. Wu, “Network reconfiguration in distribution
generated by compensation theorem. It updates system states systems for loss reduction and load balancing,” IEEE Trans. on Power
following a change in the system measurements recorded by delivery, vol. 4, no. 2, pp. 1401–1407, 1989.
micro-PMUs, without the need to rerun the whole DSSE. Case [17] R. D. Zimmerman, C. E. Murillo-Sánchez, and D. Gan, “Matpower:
A matlab power system simulation package,” Manual, Power Systems
studies confirm that the proposed method can accurately update Engineering Research Center, Ithaca NY, vol. 1, 1997.
the system states. The effectiveness of the method is scrutinized
also over system parameters uncertainties. For a reasonable

TABLE V
M ETHOD E FFECTIVENESS VS . C URRENT M EASUREMENTS U NCERTAINTY

Current Phasor Correct Neighbors ∆V Error P Error Q Error

Error SD (%) (%) (%) SD (%) SD (%) SD (%)
5 100 0 4.1 4.3 5.8
10 97.4 2.6 7.6 8.1 10.8
15 79.4 19.2 12.3 13.2 16.3
20 65 26.8 16.2 17.9 22.2
25 49.4 36 21.6 23.0 31.2