IEEE Transactions on Power Systems, Vol. 10,No.

I, February 1995 229

DISTRIBUTION SYSTEM STATE ESTIMATION

C. N. Lu J. H. Teng W.-H. E. Liu

Department of Electrical Engineering Pacific Gas and Electric Company
National Sun Yat-Sen University San Francisco, CA 94111, USA
Kaohsiung, Taiwan, ROC

-
Abstract A three-phase distribution system state estimation Nowadays, the technology to autoinatically monitor and
algorithm is proposed in this paper. Normal equation method is control a distribution system is available [141. In an automated
used to compute the real-time states of distribution systems distribution system, many meters are installed. Similar 10 those
modeled by their actual a-b-c phases. A current based formulation in the transmission system automation, real-time measurements
is introduced and compared with other formulations. are noisy, thus, techniques have to be developed to screen the
Observability analysis for the proposed distribution system state measurement data. Due to the requirements of filtering
estimation is discussed. Test results indicate that the normal measurement data and having real-lime system states for on-line
equation method is applicable to the distribution system state operation. the need for a distribution system slate estimator may
estimation and the current based rectangular form formulation is soon be justified. In this view, Wu and Neyer [IS]proposed an
suitable for this application. asynchronous dislribution system SE technique. The SE was
formulated as a equality consuained weighted least squares
-
Keywords State Estimation. Distribution Automation,
Distribution System Operation. Energy Management Systems
problem. A dismbution system SE that uses a minimum number
of remote measurements was presented in [ 161. An ilerative
procedure based on Kirchhoffs current law was used U, o w n the
1. INTRODUCTION distribution system SE.
Two types of data are required for S E they are the network
In the modern Energy Management System (EMS), State data and the measurement data. A transmission system has a
Estimation (SE) program processes a set of raw measurement relatively balanced nature which allows the decoupling of the
data and provides a real-time load flow solution which is the three phase system and the positive sequence network is used for
basis of the advanced functions for system security monitoring the analysis. In reality. power systems are unbalanced when the
and control. SE is based on the mathematical relations between power lines are not fully uansposed andor the loads are not
the system state variables (e.g. bus voltage magnitudes and balanced. The unbalanced nature of the distribution system
angles) and the measurements. Various techniques have been used prohibits the use of symmevical component transformation. and
to obtain an SE solution, excellent surveys on SE algorithms due lo various combinations of single-phase. two-phase, and
can be found in [l-31. Measurements used to compute the real- three-phase loads encountered in the distribution system. accurate
time system states are bus injections, line flows and bus study can be accomplished only with distribution lines modeled
voltages. In some estimators current magnitude and voltage angle in their actual a-b-c phase representation I5-81.
measurements are also used [4]. In addition to a real-time load In addition to the network dam, we also need a set of
flow solution, state estimator also provides functions such as bad redundant measuremenb to obtain an estimate of the system
measurement data detection, modeling error detection, meter states. In a f u l l y automated system the measurements are
placement and observability test. sufficient for SE. However, in the current stage of distribution
Real-time control of the distribution system reqbires an system automation. the number of meters installed in the system
estimate of the system stales. In the past most distribution is low and may not be sufficient for SE, i.e. the system may not
systems were not monitored, therefore, there was no need for SE. be completely observable. In order to obtain an SE under this
Under this condition, distribution system load flow program is condition, techniques have to be developed Lo provide additional
often used for planning purposes, such as in computing system data (i.e. pseudo measurements) to the estimator. Intuitively. one
losses Of different feeder configurations for system loss reduction. would use the historical dala of the feeders and distribution
Various techniques have been proposed to obtain distribution transformer loadings to provide such information. Pseudo
system load flow solutions (5-131. measuremenls estimated by using historical dala may not match
exacuy the real-time actual values. but they increase the data
9 4 WM 098-4 PWRS A paper recommended and approved redundancy of SE. If Uiis approach is adopted care must be taken
by the IEEE Power System Engineering Committee of the in assigning weights to various types of measurements.
IEEE Power Engineering Society for presentation at t h e Techniques lhai can be used to determine lhe meter or pseudo
IEEE/PES 1994 Winter Meeting, New York. New York,
January 30 - February 3, 1994. Manuscript submitted
measurement locations for oblaining a complete observability of
December 28, 1992; made available for printing the system are available in [17-211.
January 11, 1994 A three-phase SE algorithm based on normal equation
method is proposed in this paper. A new reclangular form SE
based on currenc instead of power, is introduced. The proposed
algorithm can be used U, handle many types of measuremenb. It
is applicable to the current magnitude measurements that are
often found in the distribution sysrem telemeuy. The three-phase
network models and mathematical formulation of the method are
described in this paper. The suitability of the proposed
0885-8950195/504.000 1994 IEEE

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on December 02,2023 at 19:24:55 UTC from IEEE Xplore. Restrictions apply.
 230

Bus6 Bus7 Bus8 Bus9 Bus10

15

e :Feeder Terminal Unit

:Pseudo measurement

Figure 1: A Distribution System Feeder

formulation in the unbalanced system is tested by comparing it

with polar form coupled and decoupled formulations. Va Va
Observability analysis for the distribution system SE is
discussed. Effects of measurement types and redundancy on the -Y ac ac-8
performance of estimation are investigated.
Vb Vb
2. SYSTEM MODEL AND MEASUREMENTS

Figure 1 shows a portion of a distribution system feeder in vc vc

the Taiwan Power Company (TPC) Kaohsiung District. The -Ybc-g ycc-g -Ybc-g
three-phase primary feeder operates at 22.8 KV. The circuit is
represented on a per-phase basis. As can be seen from Figure 1, Figure 3: An Equivalent Circuit of a Three-
there are three-phase and single-phase laterals involved in the Phase Feeder Section
circuit. Feeder Terminal Units (FIT0 on the feeder will be used
to collect real-time data and communicate with the master In an automated distribution system, measurements may
station. They also perform some control actions. include bus power injections, branch power flows, bus voltages
The methods developed by J. R. Carson and W. A. Lewis and line currents. However. in the present stage. the distnbution
[51 can be used to compute the impedances of circuits with system is still rarely measured. and this results in several buses
neutral and ground return paths. Line charging is ignored in the that are unobservable, i.e. only the state of a portion of the
study since it is relatively insignificant at distribution voltage system can be computed. Fonunately, historical data are available
levels. We use the approach proposed in [5-8] to build the and can be utilized to forecast the loadings of feeders and
admiWce matrix of the unbalanced three-phase system. Figure 2 distribution transformers. These data are treated as pseudo
shows the phase and neutral impedance of a three-phase feeder measurements.
section. For such a feeder section we can build a 4x4 impedance Loads in a distribution system are usually classified as
matrix including the a-b-c phases and ground node. The three general types of customers. they are the residential.
impedance matrix relates the line currents and impedances to industrial and commercial customers. Typical load pattern or
voltage drop of the line section. Figures 3 shows the equivalent daily load curve of each type of customer can be obtained by
Circuits after the elimination of the reference (ground) node in the elecuic load synlhesis or load survey technique [5.13]. The load
impedance matrix of a three-phase feeder section. The circuit in composition of each distribution transformer can be calculated
Figure 3 can be described by a 3x3 admittance matrix that according to the energy consumption of all customers served by
inc~udesthe mutual effects and relates the branch admiuances, and the transformer. B y using the load patterns and the derived load
node voltages to branch currents. The effects of neutral and composition. an hourly load of distribution transformer can be
ground return paths are accounted for in the calculation procedure. estimated and used as a pseudo measurement. The quality and
Zaa quantity of information concerning loads vary among utilities. In
Va' TPc, the energy consumption of each customer is determined by
the billing data stored in the Customer Information System (CIS)
Vb' [W.
In this study. the measurement set includes the actual
VC' measurements on the feeders and substations (e.g. branch
currents. bus voltages and branch power flows), and the pseudo
Vn measurements (e.g. distribution transformer loadings) obtained
from historical data. The weights assigned for the pseudo
Figure 2: Impedances of a Three-phase Feeder measurements are lower than those for the actual measurements.
Section In this case, the SE solution tends to match the real-time
measurements and the transformer loading data can be adjusted
accordingly. This, indeed, is an important benefit of the

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on December 02,2023 at 19:24:55 UTC from IEEE Xplore. Restrictions apply.
 23 1

distribution system SE. Based on the estimated real-time system

states, the solutions obtained from many of the distribution
system advanced application software would become more
realistic.
....,.
......
......
::::::
15
3. NORMAL EQUATION METHOD

In SE,the model used to relate the measurements and the

state variables is
2 = h(X,Y) + N (1)

W k
2 = vector of measurements
X,Y = vectors of state variables
N = measurement noise 0 5 IO 15 20 25 30 35 40 45
h = functions relating state variables to measurements Bus No.

We choose bus voltages as state variables. We can choose Figure 4: Jacobian Matrix Structure
to express bus voltages in polar form (X=O and Y=IVI) or in
rectangular form (X=IVkosO and Y=IVlsinB). N is assumed to be
a Gaussian distribution with zero mean and variance 02.u - is~ 0
used to weight each individual measurement. More accurate
measurements will have lower 0 ' s . while the pseudo 10
measurements are assigned with higher U'S to highlight the lower 20
confidence given to these measurements. The noise elements are
assumed to be independent. Let R be the covariance of N, then 0 30
z
~ i= iai2, the variance of the i-th measurement. 3 40
Weighted Least Square (WLS) estimation computes the 'E 50
>
state variable vectors X and Y which minimize the following
function 1*
wY 70
J(X,Y) = l12[Z-h(X,Y)lTR-1[Z-h(X,Y)] (2) 80
J(X,Y) is minimized by differentiating it with respect to X on
and Y,and setting the resulting nonlinear equation to zero. 'hen 0 IO 20 30 40 50 60 70 80 90
the nonlinear equation is solved iteratively by Newton's method. Slate Variable No.
Let Hi bethe measurement Jacobian matrix at the i-th iteration,
then update of the state variables can be found by solving the Figure 5: Gain Matrix Structure
following equation
Current Based SE Formulation

In the following, a new SE formulation based on curreni

Equation (3) is called the normal equation of the W S instead of power. is proposed. The proposed formulation can
problem. HiTR-lHi is called the gain matrix. A solution of handle all types of measurements. In each iteration of SE, power
measurements are converted into their equivalent currents. In
[X,Y] can be obtained by solving Equation (3) iteratively until addition, current based SE uses rectangular voltage/currenr
the vector components of the right-hand side are sufficiently coordinates. It can be shown that the gain matrix derived from
small. nodal admiaance matrix is constant.
Readers should be reminded that since we model the feeder From Equation (3) it can be seen that the gain matrix of
on a per-phase basis, each three-phase bus will have six state SE is composed of the measurement Jacobian and the covariance
variables, with two state variables for each phase. There are matrices. For a pair of phax-a power injection measurements Pa
several 3x3 blocks that locate at the diagonal and off-diagonal of
the Jacobian and gain matrices.The structures of the Jacobian and and Qa, at a bus of Figure 3. they can be expressed as :
gain matrices of the 15 bus system shown in Figure 1 are given
in Figures 4 and 5. A dot in Figures 4 and 5 indicates a nonzero
enay. Figure 4 shows the nonzero terms of JNJx.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on December 02,2023 at 19:24:55 UTC from IEEE Xplore. Restrictions apply.
 232
where Gkm + JBkm = Ykm, Gmand Bkm are 3x3 block
elements of nodal admittance matrix. Since the current injection
at phase-a of Figure 3 can be expressed in rectangular form as :

I' a = .@ s ( e a - e a . ) - b , ( f a - f , . ) + g a b ( e b - e b . )
-b ab (f b -fb,)+g,(ec-ec.)-b,(fc-fc.)l

lia = @ , ( f a - fa.)+ b,(ea- ea.)+ g a b ( f b - f b , )

+bab(eb - e b , ) + g,(f, - fc.)+ b,k, - ecJl

the Jacobian wms are

Therefore, if bus injection measurements are transformed

to equivalent currents. and bus voltages are expressed in
rectangular form, then the Jacobian matrix smcture of injection
measurements with respect to the state variables in a three bus
system are in the following form:

G31 - B31 I '32 - B32 I G 3 3 - B33

I I

B31 G31 B32 G3Z B33 '33

Where ej. fj. Ijr and Iji are 3x1 vectors. Using the same idea, if
we converl other types of measurements, such as branch power
flows and current magnitudes, into equivalent currents and
expressed them in rectangular form then the Jacobian terns of
these measurements are constant and equal to the branch
admittances.
Polar form Rectangular form With these background in hand, let's see how to find the
equivalent currents of branch power flow and bus injection
These power measurement Jacobian terms are state dependent, measurements. The branch power flow measurements P h m e a
i.e., they have to be. computed in each iteration of SE.
The Jacobian terms of bus current injections with respect and &,mea can be converted into measurement equivalent
to node voltages exp :ssedin rectangular form are as follows: current I h m e a ~ by
v the following equation:
m

where v k is the estimated bus voltage at the k-th iteration.

Current flow calculaled in each iteration is

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on December 02,2023 at 19:24:55 UTC from IEEE Xplore. Restrictions apply.
 233

Observability Analysis

When sufficient measurements are available the state

vector of the whole system can be obtained by SE. In this case
where y h is the admittance of the branch. Note that I h m e a - the network is said to be observable. This is m e when the rank
W' and &d are complex values and expressed in rectangular of measurement Jacobian matrix is equal to the number of
forms. unknown slate variables. The rank of the measurement Jacobian
The conversion of bus injection measurements to tbeir matrix is dependent on the locations and types of available
equivalent currents is the same. measurements as well as on the network topology. In the three-
phase formulation, the rank may also be affected by the coupling
terms between phases. In certain cases, one phase may be
= ((P+jQ)kmea / vk)* observable while the others are not. Thus, the relationship
= h(Ikmea-eqv) + j h(Ikmea*qv) (7) between numerical observability [ZO] and topological
observability [21] is not obvious under these conditions. It Seems
Where (P+jQ)kmea are bus injection measurements at bus k. The not very straightfonuard to define topological observability for
three-phase SE formulation. Consequently. a numerical
bus injection calculated at each itelation is observability approach m a y be preferable for the proposed three-
phase SE.
Ikd = z The numerical observability analysis based on viangular
= Re&d) + J h(Ikcal) (8) factorization of the gain matrix can be applied to the three-phase
SE without major moditications. If any zero pivot is encountered
In addition to the power measurements, other types of during the factorization of the gain malrix. it indicates that the
measurements such as current and voltage magnitude corresponding sme variable is not observable. In the three-phase
measurements can also be used in the rectangular formulation. formulation, the zero pivot may correspond to one specific phase
For the current and voltage magnitude measurements, due to the of a bus. This numerical observability algorithm can be extended
lack of phase information, we use a slightly different procedure to to suggest additional meter placement.
obtain their equivalent complex currents and voltages. The
equivalent currents and voltages are equal to the measured 4. TEST RESULTS AND DISCUSSIONS
magnitude values (IVklmea, IIhlmea) multiplied by the ratio of
The described distribution system SE formulation has been
the ca~culatedcomplex values (Vkcal = lVkCall LeV, ~ h c a=l tested by using three feeders in the TPC disvibution system.
I I ~ LeI)~ Ito the magnitudes of the calculated values (IVkcall, Table 1 shows the peninent data of these feeders. Note tha each
bus of the feeder involves 3 single-phase pseudo measurements
I I ~ ~ Ii.e.,
) , let
and 3 single-phase line flow or current magnitude real-Lime
measurements if there is an Flll at the bus. The average rlx ratio
~ h c a =l [(P+jQ)hCa' / vkcal]* (9) of the line is 1.88. The impedance and admimce matrices, given
in ohm/mile, are as follows:
then
2.7834 + 11.4794 0.2352 + 10.6561 0.2334 + j0.5827
1 b r n e - P = lIbmea I (1hCal I IIhcall)

ad
= Re(Ihmea-W) + j h(1hmea-W') (10)
0.2352+p.6561 2.7871 +jl.4761 0.2352+p.6561
0.2334 +j0.5827 0.2352 + j0.6561 2.7834 + 11.4794

0.2951 -j0.1297 -0.0597 -j0.0136 -0.0538 -p.0071

1
-0.0597 -p.0136 0.2956 -j0.1261 -0.0598 -10.0136
v k r n a q v = lvkmea I (vkcal / Ivkcall) -0.0538 -p.0071 -0.0598 -j0.0136 0.2952-p.1298
= Re(Vkmeaeqv) + J h(Vkmea-eqv) (11) Table 1: Measurement and Feeder Data
By using Equations (5). (7). (9) and (IO) we can convert
power measurements and current magnitude measurements into
their equivalent currents. Based on equivalent currents the
Jacobian tern are constant and equal to the admittance matrix
elements. The equivalent currents of the measurements 1mea-W.
1 I 1 1 I
Feeder
No.
No. of
Buses
Length No.of
FlU's
I
No. of
Pseudo
Measure-
mentS
~

I
No. of
Actual
Measure-
rnents
1
calculated currents IC', measurement Jacobian matrix H, and the
covariance matrix R are used to compute the right-hand side
vector of Equation (3). In addition to power and current
measurements, the voltage magnitude measurements, Equation 93
(11). can also be used. The first derivative of equivalent voltage 48 12
expressed in rectangular form with respect to the state variable is 48 18
unity. Based on constant Jacobian matrix H, the gain matrix
HTR-lH is constant, and it needs to be built and factorized only
once. Great improvement in the SE execution time can be
obtained by using this new scheme. Figure 7 shows the flow
diagram of the proposed three-phase SE algorithm.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on December 02,2023 at 19:24:55 UTC from IEEE Xplore. Restrictions apply.
 234
SE formulation (method 3) has the best performance in all tests
Estlmatc pseudo measurcmcnts conducted. Iteration number required by the decoupled version is
at distnbuuon bansformers
higher. Test results indicate that the decoupled formulation has
& difficulty in handling current magnitude measurements. Fully
coupled formulation is more stable than the decoupled
Assign weights for pseudo formulation. It is also shown in Table 2 that if the measurement
mCaSurcments
set includes current magnitude measurements, all three
4 formulations require a higher execution time to obtain the
solution. This is due to the lack of phase information in the
Use equations (5).(7).(9),(IO).( I t ) IO
convert mcaswmcnts data into current magnitude measurements.
equivalent mcajurements expressed in During the testing of the proposed method, it was found
rectangular form that if the current magnitude measurements that have high
Jm weights are involved. a special procedure should be followed
when a flat s m is used. From equations (9) and (10). i t can be
Compute and factorize the gain seen that the equivalent complex currents obmned from a flat
matrix
start would be far away from the solution. this might result in a
convergence to an incorrect solution. To resolve this problem, a
tentative solution obtained by greally reducing the weights of
current magnitude measuements in the first and second ilerations
nght-hand side vector of equauon (3) of SE, is found first. After that the weights of current magnitude
measurements are restored back to their original values, and the
+ solution procedure continues. For instance. in Table 3, for the
case of 60 bus system with current measurements, i t requires two
Solve equation (3) and update (X,Y) iterations to obtain the tentative solution and another three
iterations to get the final solution. In this case. the gain matrix
needs to be built and factonzed twice. Using this procedure. it
seems that method 3 handles current magnitude measurements
more effectively than other methods. The above mentioned
procedure is not required if no current magnitude measurement is
involved or when a previous SE solution is available.
Table 4 shows the final J(X) values of the reponed cases.

A,
Since method 3 has converted the power measurements into the
equivalent current measurements. the equation used to calculate
the J(x) is redefned as:

Figure 7: Flow Diagram or the Proposed Algorithm

The measurement data are simulated by using a three-phase The a's used in the above cquauon are computed from a's of real
load flow solution. Noise is added to each measurement and and reacuve power measurements
weights of measurements are given. Noise is added randomly to Table 5 shows the deviauons of the estimated solution
the actual and pseudo measurements and is at the ranges of *IO% from the load flou solution which is used to probide the
and f30% respectively. Two different weights are given to the measuremen1dam The demauon is defined as
measurements, one for the actual measurementsand the other one
for the pseudo measurements. The weights are 1/3 and 1/50
respectively.
Three different methods are tested, they are :
Method 1 : Fully coupled version of normal equation method
expressed in polar form.
Method 2 : Decoupled version of normal equation method
expressed in polar form. where PkmSe is Ihe estimated branch flow and pkmload is
Method 3 :The proposed current based method. the corresponding value in the load flow solution. I t should be
In method 2, tbe gain matrix is approximated by a block reminded that since measurement noise is included randomly iii
diagonal matrix, i. e. ignoring the off-diagonal terms to get the the measurements. 2 1070. and 230% respectively for actual and
decoupled gain matrices correspondingto real and reactive power, pseudo measurements, thc final SE solutions will not be the
thus, decoupling the angle and voltage components. Various tests same as the load flow solution. The results shown in Table 5
which use different types and numbers of measurements are indicate that even with only the pseudo measuremen& the
conducted. Some of them are reported here. Tests are conducted proposed algorithm can provide an S E solution. However, due to
on a SUN IPC workstation. Tables 2 and 3 show the normalized the low accuracy of me pseudo measurements. the solution is bad
execution time and number of iterations required in each case.NC when we compare i t with the base case load flow solution. This
indicates the non-converged case. In addition to voltage estimated solution can be improved by adding real-time
magnitude measurements, two types of actual measurements are measurements to the measurement set. As can be seen from
tested, they are the power flow measurements and iurrcnt Table 5, cases with higher number of actual measurements have
magnitude measurements. The execution time of method 3 with lower deviations, i.e.. they match more closely with the assumed
P, Q measurements is used as a reference and normalized to 1. real-time solution. Test results have indicated that when current
It can be seen from Tables 2 and 3 that the current based

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on December 02,2023 at 19:24:55 UTC from IEEE Xplore. Restrictions apply.
 magnitude measurements are involved, due to the lack of phase
information. the deviations are higher, but the percentage
differences between different methods are still small.
To verify that all three formulations have the simi!ar
solutions, using the solution obtained from method 1 as a
reference. we compute the summation of differences in branch
flows of solutions obtained from different methods. It was found
that the differences of the case of 60 bus system with only
pseudo measurements are within 1.1%. When the real-time
measurements are added. the deviations are within 0.4%. This
result indicates that the three tested methods did not converge
exactly to the same solution, but they have very similar
solutions.
With the aid of the actual measurements the proposed
distribution system SE is able to provide a load flow solution for
real-time operation. An accurate real-time update of the bus load
forecast can also be performed by using SE solution.
It can be seen from Figure 5 that the gain mamx is sparse,
thus, sparse maUix technique should be used in solving the
problem. Since individual feeder SE can be handled separately and
the problem size is generally small, the distribution system SE Soluliaos wtlb only p c u d o Solutlws wilh re~l-lmmcand pseudo
can be performed on smaller machines or processed mesiurernenli measuremenis
Method M e l b o d M e ~ D o d M i l h o d Mclbod Melhod Number
simultaneously by different machines. The tests are also 1 2 3 I 2 3 OlFrLIS
conducted on a 486 PC and similar results have been obtained. 12.514 14 I14 13.584 I82S8 2.224% 2.195% 5
% e %
13 SU4 11126 2.1WB 2.0378 10
Table 2: Normalized Execution Time 12.514
e
14 I 1 4
e e

-mcnts
p,~' 1.900 1.450 1.000 4
15 BUS Plows 1.900 1.450 1.050 6
ancal+ 3.050 PE 2.150 4
3.100 PE 2.150 6
Execution PO' 2.071 1.119 1.000 5

N0.M Actual Metbod Method Metbod N0.M

bU6CS measure 1 2 3 Fmr

No. of
Ilcralioas 7. REFERENCES

F. F. Wu. "Power System State Estimation: A Survey,"

60 Bus Rows 2 I IO I 2 10 Electrical Power & Energy Systems, Vol. 12, No. 2,
cumcot+ 6 I NI 5 5 April 1990.
6 1 K I 5 IO M. B. Do Coutto Filho. A. M. Leite da Silva, and D. M.
Falcao, "Bibliography on power system state estimation
(1968-1989)," lEEE Trans. on Power Systems. Vol. 5,
No.3 Aug. 1990.
A. Bose, and K . Clements. "Real-Time Modeling of
Power Networks." IEEE Proceedings, pp. 1607-1622.
Dec. 1987.
K. I. Geisler, "Ampere Magnitude Line Measurements for
Power System Stare Estimation," IEEE Trans. on Power
Apparatus and Systems, Vol. PAS-103, No. 8, Aug.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on December 02,2023 at 19:24:55 UTC from IEEE Xplore. Restrictions apply.
 1984. "State Estimator Measurement System Reliability
D. I. H. Sun, Distribution System Loss Analysis and Evaluation - An Efficient Algorithm Based on Topological
Optimal planning, Ph. D. Dissertation, University of Observability Theory," IEEE Trans. on Power Apparatus
Texas at Arlington, May 1980. and Systems, April 1982, pp. 997-1004.
D. I. H. Sun, S. Abe, R. R. Shoults, M. S . Chen, P.
Eichenberger, D. Farris, "Calculation of Energy Losses in Chan-Nan Lu received B.S. degree from National Taiwan
a Distribution System," IEEE Trans. on Power Apparatus University, M.S. degree from Rensselaer Polytechnic Institute,
and Systems, Vol. PAS-90, No. 4, pp. 1347-1356, and Ph. D. degree from Purdue University. He has held positions
July/Aug 1980. at General Electric Co. Pittsfield. Mass., and Harris Corp.
N. Vempati, R. R. Shoults. M. S . Chen, L Schwobel, Controls and Composition Division, Melbourne, FI.. Currently.
"SimplifiedFeeder Modeling for Load Flow Calculations," he is with Department of Electrical Engineering. National Sun
IEEE Trans. on Power Systems, Vol. PWRS-2, No. 1, Yat-Sen University, Kaohsiung, Taiwan. He is a senior member
Feb. 1987, pp. 168-174. of IEEE.
T.-H. Chen, M.-S. Chen K.-J. Hwang, P. Kotas, and E.
A. Chebli, "Distribution System Power Flow Analysis - J. H. Teng received his BS and MS degrees froin National Sun
A Rigid Approach," LEE Trans. on Power Delivery, Vol. Yat-Sen University in 1991 and 1993. Currently, he is pursuing
,NO. 3, July 1981, pp.1146-1152. his PH. D degree in the same University.
K. A. Birt, J. J . Graffy, J. D. McDonald, and A. H. El-
Abiad," Three phase load flow program," IEEE Trans. on W.-H. Edwin I.iu received the B.S. degree from National
Power Apparatus and Systems, Vol. PAS-95, No. 1, Taiwan University in 1981, the M.S. degree in 1984 and Ph.D.
Jan/Feb 1976, pp. 59-65. degree in 1987 bo& from the University of California Berkeley,
M. E. Baran, F. F. Wu, "Network Reconfiguration in in Electrical Engineering and Computer Sciences. He was a
Distribution System for Loss Reduction and Load research assistant at U. C. Berkeley from 1983 to 1987 and
Balancing," IEEE Trans. on Power Delivery, Vol. 4, No. worked for Bonneville Power Administration during the Summer
2, pp. 1401-1407, 1989. of 1986. From September 1987 to June 1991, he worked for
C. C. Liu, S. 1. Lee, K. Vu, "Loss Minimization of Empros Systems International as a Senior Engineer in the
Distribution Feeder: Optimality and Algorithm," IEEE Network Applica*ion group and developed solfware for power
Trans. on Power Delivery, Vol. 4, No. 2, pp. 1281-1289. system computer applications. Since J u l y 1991, he has been
1989. with the Applications and Systems Integration Department of
H. D. Chiang. R. J. Jumeau, "Optimal Network PG&E where he is respondible for several research and
Reconfigurations in Distribution Systems: Pan 1: a New development projecls in both analytical melhodology and
Formulation and a Solution Methodology," IEEE Trans. computer applicauons. Dr. Liu also leaches graduate courses in
on Power Delivery, Vol. 5, No. 4. pp. 1902-1909, 1990. the Engineering Division of San Francisco S w e University.
C. S. Chen, and M. Y. Cho, "Determination of Critical
Switches in Distribution System,'' IEEE Trans. on Power
Delivery, Vol. 7, No. 3, July 1992. pp. 143-1449,
"Distribution Automation: A practical tool for shspinp a
more profitable future," Special Report, Electrical World,
Dec. 1986, pp. 43-50.
F. F. Wu and A. F. Neyer, "Asynchronous Distributed
State Estimation for Power Distribution Systems."
Proceeding of 10th Power Systems Computation
Conference, Aug. 1990.
I. Roytelman, and S. M. Shahidehpour, "State Estimation
for Electric Power Distribution Systems in Quasi Real-
Time Conditions," EEE paper 93 WM 090-1 PWRD.
F. Mafaakher, A. Brameller, J. F. Bermudez, "Optimum
metering design using fast decoupled state estimator,"
IEEE Trans. on Power Apparatus and Systems, Vol. PAS-
98, Jan. 1979, pp. 62-68.
K. A. Clement, G. R. Krumpholz, P. W. Davis, "Power
system slate estimation with measurement deficiency: an
observability measurement placement algorithm," IEEE
Trans. on Power Apparatus and Systems, Vol. PAS-102,
July, 1983, pp. 2012-2020.
M. K. Celik, and W.-H. E. Liu, "A Meter Placement
Algorithm for the Enhancement of State Estimation
Function in an Energy Management System," Paper
prepared for the Third International Symposium on
Electricity Distribution and Energy Management, 1993.
A. Monticelli and F. F. Wu, "Network Observability:
Identification of Observable Islands and Measurement
Placement," IEEE Trans. on Power Apparatus and
Systems, Vol. PAS-104, May 1985, pp. 1035-1041.
K. A. Clement, G. R. Krumpholz, and P. W. Davis,

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on December 02,2023 at 19:24:55 UTC from IEEE Xplore. Restrictions apply.
 237

DISCUSSION of the transmission systems weighted least square state estimator

method to distribution systems. Distribution systems control and
SLUTSKER, Siemens Energy Br Automation, Plymouth, operation have fundarnedtal differences with t h k e of transmission
Minnesota. W.F. TINNEY. Consultant, podand, Oregon: The authors systems which make it diffcult to apply the transmission Systems state
are congratulated on the development of a new method of Slate estimation techniques directly to distrihution systems. The differences
estimation for distribution svstems. we would like to s o k i t include unbalanced conditions, current magnitude measurements and
the applications of historical (statistical) load data as pseudo
measurements. The authors have proposed reasonable ideas and
In the authors' method. power flow measurements are translated proved by test results that the p r o p i d ideas work. We have a few
into equivalent current values that are then used in the state questions and comments regarding the practical implementation of the
estimator solution. While it is easy to convert power measurements proposed method and presented results.
into currents. it is much less clear how to compute weights of current
measurements to ensure that the solutions of the original and
derived problems are identical. 1. Modeling of unbalanced distribution systems:
It can be shown that, if real and reactive power measurements The authors have based their algorithm on three wire presentation of
have different weights, the error terms in real and imaginary four wire systems. The elimination of neutral wire may he. done
components of the derived currents are distributed normally with according to [a], using one of the following two methods. the first one,
variances which are functions of voltage components used in current refereed to as the Kron reduction method, makes an assumption that
computation. This means that weights of current measurements, the ground wire is at zero potential at both ends of a branch (feeder
which must be computed as inverses of variances to guarantee the section), or,
equivalence between the original and derived formulation. will
fluctuate between iterations and the gain matrix will no longer be
constant.. The need to recompute and refactorize the gain matrix in
zi,* znj I ,z
each iteration will make the proposed method no more advantageous
than the fully coupled formulation. Of course, the gain matrix can be
held constant at some value of current measurement weights, and The second method, refereed to as the neutral return reduction
this will produce a solution but it won't be the solution of the original method, makes the assumption that the return current follows the
problem. path through the neutral and all loads are grounded, or,
It appears that the authors used measurements with equal
weights of real and reactive components in their tests. This,
however, will not be true in many practical situations. How do the
authors intend to deal with cases bf unequal weights of real and
reactive measurements? Can the authors explain their approach to
the calculation of weights of the derived currents?
These have Opposite impacts On the
(1) decreases self impedances slightly and decreases mutual
muation
imoedances simificantlv. Euuation 121 increases both series and
Again, the authors are commended on a new method of solving a m;tual impedances. Hdwevir, some bi the simplifying assumptions
state estimation problem. Their answers to our questions will be may not work for practical Cases. Neutral wire may not always
awaited with great interest. represent zero potential, and there are delta connected loads in every
distribution feeder. In general, unbalanced 4-wire distribution systems
Manuscnpl received February 17, 1994 should be simulated as 4 wires without elimination of neutral wire
using a 4x4 matrix

2. Branch current magnitude measurements:

Wayne Hong Opercon Systems, Inc :
The problem with current magnitude measurement is that the only
The authors are to be commended for their interesting development of DSE. This
way we can obtain current angle is to use preliminary power flow
paper presented a thorough mathematical framework and described some potential results based on load pseudo measurements. Our experience with this
applications, including those for real-time operation. I would appreciate if the method indicates that the weight coefficients for current magnetite
authors can comment on the following: and angle should be different as magnitude is directly measured and
is more reliable, however, angle is estimated using historical data.
I ) A separate paper [AI in the '94 Winter Meeting also described DSE. How do Another problem with weights for current magnitude is that they are
these two paper differ in their approaches and applications? less than weights for active and reactive power measurements in
distribution substations, as the corresponding power is determined by
2 ) In most distribution systems, there are never enough real-time measurements. As state estimation in transmission systems.
such, the networks are in general not observable. You have suggested to use
historical data as pseudo measurements. Could you quantify the relationship
between mors in the historical data and those in the resulted system states obtained 3. Weights for pseudo measurements:
from DSE?
The authors use different weights for real and pseudo measurements.
3) According to your paper, a DSE with added real-time readings can provide a But for different type pseudo measurements different values of
better estimate of the state of dishibution system operation. How does DSE then fit weights are to he used. Another observation is that loads (pseudo
into the overall goal of Distribution Automation? measurements) power factor is more reliable than active power or
current magnitude. The power factor is used as a measure of results
[AI M.E.Baraa A.W. Kelley, "State Estimation for real-time Monitoring of
(loads) feasibility and should be kept within its permissible range [b].
Disnibution Systems". 94 W M 235-2 PWRS

Manuscnpl received FCbruary 24, 1994 4. Proposed simplifications:

The authors assume that the phase angles of voltages are less than 10
degrees. Our experience with distribution systems indicates that this
assumption is correct only for the primary voltage level and in cable
I. Roytelman (SiemensEmprus) and S.M. Shahidehpour (Illinois systems. Since loads are connected to the secondary voltage level,
Institute of Technology)-- The topic of this paper is extremely distribution transformers voltage drop should be. taken into account.
important from the standpoint of future development of distribution In heavily loaded overhead lines, line voltage drop gives additional
management systems. The authors have proposed the implementation angle shift. These assumptions may provide a more simple model for

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on December 02,2023 at 19:24:55 UTC from IEEE Xplore. Restrictions apply.
 238

distribution systems, however, the simulation results may turn out to Let
be different from actual practice.

5. Test results:

Test results are presented for relatively small systems. To our

knowledge, it is better to apply state estimation to portions which are
supplied from distribution substation transformers rather than feeders.
The reason is that p e r flows are measured at substation
transformers and current magnitudes are measured at corresponding
feeders. O n the average, one feeder supplies 150-200distribution Assuming that the complex voltages used in the
transformers and the number of feeders connected to one substation conversion are deterministic and the active and
transformer is 7-10. So the average size of a portion entails to reactive power measurements are statistically
1500-2000transformers. If we consider unbalanced conditions as well,
the state estimation problem will be a complicated problem for independent, then, the variances of the converted
distribution systems and it would be interesting to see the results of complex currents can be derived as follow:
the proposed techniques on more practical cases.
The authors should be congratulated for a well-written paper on Var(Re(Iimea-eqv)) = Var(aPimea+bQjmea)
distribution systems.
= a2Var(Pimea) + b2Var(Qimea) ('44)
[a] IEEE Tutorial on Power Distribution Planning, 92 E H 0 361-6-
PWR.
Var(Im(Iimea-eqv)) = Var(bPimea-aQimea)
p] I. Roytelman and S.M. Shahidehpour, "Practical Aspects of
Distribution Automation in Normal and Emergency Conditions," IEEE
Trans. on Power Delivery, Vol. 8, No. 4,pp. 2002-2008,October 1993
= b2Var(Pimea) + a2Var(Qimea) (A3

Manuscript receivrd March 4, 1994. In the case where Var(P) = Var(Q), the equations can
be simplified as

Var (Re(Iimea-eqv)) = Var(Pi)Ni2 = Var(Pi) (A6)

VW (Im(Iimea-eqv)) = Vw(Qi)Ni2 = Var(Qi) (A7)
C. N. Lu, J. H. Teng and W. H. E. Liu : W e wish
to thank the discussants for their interest in our paper In our implementation, by using a distribution system
and thoughtful comments. In the following, we will load flow or a previous state estimation solution, an
respond to each of the discussants' comments. initial set of e and f is obtained. This set of e and f is
then used to compute the variances of equivalent
Messrs Slutsker and Tinney measurements and form the gain matrix. The gain
matrix is kept constant while the right hand side of
In order to obtain a constant Jacobian matrix and to equation (3) in the paper are computed by the solution
incorporate current measurements in the formulation, from the previous iteration including the weights. Note
all measurements are converted into equivalent current that the weights for the measurements are set to be
measurements. The advantage of the formulation has equal to the inverse of the variances.
been clearly shown in the test results of the paper.
Strictly speaking, the voltages used in the calculation of
The conversion process is explained in further detail in the right-hand-side vector of equation (3) do carry
the following. As we have stated in the paper, at each certain randomness through the iterative process.
iteration, the equivalent current measurements are However such randomness is neglected in the
calculated based on the active and reactive power traditional WLS state estimation. The variance of each
measurements and the solved voltages from the component of the vector [z-h(xi,yi)] is assumed to be
previous iteration.
equal to the variance of the corresponding component
of z, i.e. Var(Zk-hk(Xi,Yi))'Va(Zk). This is based on the
assumption that xi and yi from the previous iteration are
deterministic, and hence h(xi,yi) is deterministic. Such
assumption has been widely accepted in practical
implementation. Similarly, in our derivation of equivalent
measurement variances, the voltages are considered as
deterministic. Statistically, with the approximated
variances, the estimator is still an unbiased estimator.
Practically, we think, it is not necessary to get into
detailed derivation for the randomness in the nonlinear

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on December 02,2023 at 19:24:55 UTC from IEEE Xplore. Restrictions apply.
 239

iterative process. Based on our experience, slight 3. State estimation can provide a good estimate of the
approximation in the variances (i.e. weights) does not system states that can then be used by many
affect the estimation solutions too much. In order to distribution automation applications, such as feeder
validate the approximation, a 15-bus system example is reconfiguration, v o l t a g e h a r control, etc.. T h e
tested under various conditions, and results are shown usefulness of these applications requires a reasonable
in the following Table. accurate estimate of the system states. As can be seen
~~~~ ~~ ~~~ ~
in the paper, the quality of the forecasted load data are
-Weights Method 1 Method 3 improved by the distribution state estimation using
est
-15.7151
P5a,mea- p=2
Q=2
P~~ ~ 14.20619 P5aest = 14.2041 available real-time measurements. The state estimation
est
Q5a = 2.45973 Q5= est = 2.47091 can be used to check the validity of the forecasted load
p5= mea--3.10072 P=20 P5aest = 14.70647 PSaest - 14.7011
~
and to make necessary corrections. Hence, the system
Q.20
QSa
est
= 2.67225 Qsaest = 2.68437
states provided by state estimation is better than what
is available from an off-line power flow study.
P.200 Psa est 15.48135 P5aest =15.47671
Q=20
=
Furthermore, if more real-time measurements become
est
Q5a = 2.67230 Q5aest =2.68651 available in some key locations and the distribution
network becomes observable, the system states can be
mea: measurement estimated directly from the real-time data. This can
est: estimated value highly improve the solutions of all network applications.
A fully automated distribution system may become
Methods 1 and 3 correspond to the fully coupled polar feasible.
form method which uses the original measurements
and weights, and the proposed current based method,
respectively. Method 3 uses the above mentioned Drs Roytelman and Shahidehpour
method to determine the weights of equivalent
measurements. As can be seen from the Table, for 1. We agree that different model should be used if the
both methods, when the original weights are increased line under study does not have a multi-grounded neutral
(variances are decreased) the estimated injections which is the basic assumption of the Kron reduction
match the measurements more closely, and their method adopted in this paper.
results are very similar. Resembling results are
observed at different phases and nodes of the feeder. 2. In our method the current magnitude measurements
These results indicate that the measurement and are treated as regular measurements except for the
weight conversion is properly handled by the proposed special treatment in a flat start case. The estimated
method. phase angles of the corresponding branch currents are
obtained at each iteration of the estimation process. W e
Mr. Wayne Hong did not use current phase angle measurements in the
estimation.
1. Both [A] and our paper use weighted least square
technique and a three-phase model to estimate the 3. U n d e r t h e c i r c u m s t a n c e s that r e a l - t i m e
distribution system states. The major difference measurements are not sufficient for a state estimation,
between these two papers is in the approach used to any available historical data related to the feeder
obtain the constant gain matrix. In [A], some loading should be used as pseudo measurements.
approximations and transformation are used to simplify Certainly, due to different confidence on various
the Jacobian matrix while in our method, constant histoiical data, different weights are given to different
Jacobian matrix is obtained by converting power types of data. Since weight disparity has beeil identified
measurements into equivalent currents and expressing as one of the factors that could cause numerical
bus voltages in rectangular form. Both methods are problems in state estimation. The effects of weight
designed to provide an estimate of the real-time system assignment should be carefully investigated. In this
states for distribution system operation. paper, we do not explicitly address the issue of weight
assignment. The purpose of the paper is to present an
2. Due to the cost of measurement and communication effective formulation to solve distribution system state
equipments, real-time load data of each node in the estimation. Further research and development in many
feeder are not available at the present time. Load data important subjects in this area, such as meter
are often obtained based on historical data collected by placement, observability analysis, bad data processing,
utilities. Since these data are used in the estimation weights selection, etc., are certainly welcome.
process, thus, the accuracy of load data affects the
estimation of the real-time system states. Based on 4. In this paper we did not assume that the phase
some assumptions on the measurements and load data, angles of voltages are less than 10 degrees.
Table 5 provides some quantitative results of the errors
existing in the historical load data and the estimated 5. To be useful in the feeder automation, we tested the
results. proposed method on individual feeder. Since data

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on December 02,2023 at 19:24:55 UTC from IEEE Xplore. Restrictions apply.
 240

accuracy and sufficiency of different feeders may be

different, the strategy used in this paper is to estimate
the states of all feeders sequentially one after another.
Therefore, the test system size is relatively small. If
sufficient data are available then a complete state
estimation covering all feeders simultaneously can be
performed.

Once again the authors would like to thank all the

discussants for their comments and contributions.

Manuscript received April 29, 1994

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on December 02,2023 at 19:24:55 UTC from IEEE Xplore. Restrictions apply.