Sunteți pe pagina 1din 4

IEEE T&D Asia 2009 1

Topology Error Identification Using Branch Current


State Estimation for Distribution Systems
Mesut E. Baran Jaesung Jung Thomas E. McDermott
Dept. of ECE, NC State University EnerNex Corp. Pittsburgh, PA


Abstract The branch-current-based state estimation (BCSE) - Topology processor
method is a novel approach to solve distribution state estimation - Observability analysis
(SE). However, without suitable topology error processing for - State estimator
BCSE, the value of the BCSE will be degraded in real world
- Bad data processing
applications. Therefore, this paper introduces a method for
topology error identification based on the use of normalized Although the BCSE method is a novel approach to solve
residuals resulting from the employment of the BCSE method. distribution SE, without incorporation of these procedures, the
The effectiveness of the proposed approach is demonstrated and value of the BCSE method will be seriously degraded for
in addition, the topology error identification results obtained by practical application. Accordingly, integration of these
using the proposed approach are presented in this paper. procedures into the BCSE method is desirable. Hence, in this
paper, we first show topology error identification based on the
Index Terms Topology Error Identification, Power System
State Estimation, Power Distribution System.
use of normalized residuals from the result of BCSE method.

I. INTRODUCTION II. BRANCH-CURRENT-BASED STATE ESTIMATION

D istribution state estimation (SE) has a critical role in the The BCSE method [2] was developed based on the
Distribution Management System (DMS) for the weighted least square (WLS) approach and uses the branch
currents as the state, i.e.
estimation of unknown states using limited measurement
information. For reliable and optimal DMS control, several SE
methods have been proposed. There are two typical xi [ I ri , I xi ] (1)
approaches to SE. The first includes algorithms based on where I ri is branch current real part and I xi is branch current
power flow [1, 2], the second includes extensions of the
imaginary part. Hence, BCSE estimates system state x by
conventional SE for three phase analysis [3, 4]. Although the
solving following WLS problem :
SE approach is preferred over the power flow approach, its m
2
computational complexity may prevent its use in practical f min ( J ( x )) wi ( zi  hi ( x ))
i 1
applications.
In this paper, the Branch-current-based SE (BCSE) method T
[ z  h( x )] W [ z  h ( x )] 2
[2] is considered, since it is very suitable for distribution SE.
This method shows excellent performance in terms of where wi and hi ( x ) represent the weight and the
computation speed and memory requirements. Furthermore, measurements function associated with measurement zi
the main advantage of the BCSE is that the measurement
respectively. For the solution of this problem the conventional
functions are simplified for power and current measurements iterative method is adapted by solving following the normal
are taken from an unbalanced radial distribution feeder by equations at each iteration to compute the update xk 1 xk +xk
using the branch currents as the state.
SE procedure became one of the most frequently used k k T k k
power system applications in real world employment. There [G ( x )]'x H ( x )W [ z  h( x )] (3)
are various issues that need to be addressed for the SE to be where
successfully implemented in practices [12]. In this broader G ( x)
T
H ( x )WH ( x )
sense, SE can include :
is the gain matrix and H is the Jacobian of the measurement
function h ( x ) . Hence the only difference between the node
This work was supported in part by the U.S. Department of Energy under
Grant DE-FG02-06ER84647. voltage based SE and BCSE is the measurement functions
M. Baran is with Department of ECE, NC State University, Raleigh, NC associated with the type of measurements to be processed.
USA (e-mail: baran@eos.ncsu.edu).
J. Jung is a graduate student in the Dept. of ECE at NC State University in
Raleigh, NC (email: jsjung@ncsu.edu)
T. E. McDermott is with EnerNex Corp. Pittsburgh, PA, 15236 (email:
tom@enernex.com)
2

III. TOPOLOGY ERROR IDENTIFICATION where R '


1 T
( R  HG H ) and reevaluate the state estimates
A. Proposed Method along with the sum of squares f . If the new value of f
SE relies on the basic assumption that we know the exact satisfies the chi-square test of inequality, then the omitted
network model so that we can write the measurement measurement has been successfully identified as the bad data
functions h ( x ) beyond any doubt. However, in the practical point.
world, the status of switching devices is unknown or, for some C. Topology Error Processing
reason, the current values in the database are under suspicion.
In this topology error detecting method, changing the on/off
Hence, the risk of assuming the wrong status for the switching status of branches one after the other is done and then state
device will not be completely avoided, potentially leading to estimation is performed in each case. If after reversing a
topology errors. Topology errors usually cause the state branch status, residues in one of the estimation runs are within
estimate to be significantly biased. As a result, the bad data threshold values then the original branch status is declared
detection & identification routine may erroneously eliminate false. Based on proposed method, the basic flow is illustrated
several analog measurements which appear as interacting bad in Fig. 1.
data, yielding an unacceptable state. Therefore there is a need
to develop effective mechanisms intended to detect and
identify this kind of gross error. The intention of this method
is to present approaches that deal with topological errors and
related matters using a bad data detection & identification
routine.
B. Bad Data Detection & Identification
When the system model is correct and the measurements
are accurate, there is good reason to accept the state estimates f  F k2,D
calculated by the WLS approach. But if a measurement is
grossly erroneous or bad, it should be detected and then
identified so that it can be removed from the estimator
calculations. This is called bad data detection & identification
in state estimation procedure [11]. The bad data detection
procedure is described below :
Step1 - Calculate the estimated errors e j z j  z j after
running BCSE where z j and z j represent the measurement
and the estimate at j-th node respectively.
Step2 - Evaluate the weighted sum of squares, called
inequality
Nm
f e j / V j
2 2
(4)
j 1

Step3 - For the specified probability D and the appropriate


number of degrees of freedom k ( N m  N S ) where N m and
Figure 1. Flowchart of Topology Error Processing
N S represent number of measurements and state respectively
variable, determine whether the value of f is less than the Step1 - Detect a topology error using bad data detection &
critical value corresponding to D or not. In practice, this identification. When a topology error occurs, it will be
detected in the weighted sum of squares, an inequality.
means we check that the inequality satisfies the following
Step2 When a topology error is detected, change the switch
condition :
on/off status. This change is performed sequentially by switch
f  F
2
k ,D (5) information in input data.
If so, then the measurements and the state estimates are Step3 - Perform the BCSE using the input data which changes
accepted as accurate. the switch on/off status.
Step4 - Run bad data detection & identification using the
Step4 - when the requirement of inequality is not met, there is
BCSE result and save the inequality value in every case.
reason to suspect the presence of at least one bad data
Step5 - If the branch having switch is end, go next step
measurement. Upon such detection, omit the measurement
otherwise, go step 1.
corresponding to the largest standardized error, namely, Step6 - Define which switch status is wrong. Using the saved
'
( z j  z j ) / R jj ( 6) inequality value, the smallest value represents no topology
3

error. So the switch status having smallest inequality value is distribution with a standard deviation V of 0.0233 (accuracy is
correct but current known status is wrong. 7% of their measured values). The voltage measurement data
are generated by adding measurement error with a standard
IV. TEST RESULTS deviation V of 0.0067 (2% measurement error).
For topology error processing, a test feeder is used. The For bad data detecting & identification, the threshold value
test feeder is an IEEE 34 node radial test feeder with 23 kV for inequality is chosen as D 0.05 . Case 1 has 111
three-phase primary [5]. A reduced version of this test feeder measurements ( N m ) and 102 state variables ( N S ) which has
is used to facilitate debugging and assessment. A one-line
diagram of the reduced feeder is given in Fig. 2. For test 9 degrees of freedom ( k ) so the threshold value is set at
2
purposes, distributed line section loads are lumped equally at F 9,0.05 16.92 for case 1. For case 2, it has 9 more
terminal nodes of the line sections. The nominal load data is measurements than case 1 so the threshold value is set
used as actual load which is determined by power flow results 2
as the correct measurement. The feeder is modeled to have a at F18,0.05 28.87 for case 2.
heavy loading condition on feeder and the line data is used
given in [6]. The total load is x Switch Error Detection
S load , a 490  j 315 kVA For testing switch error detection, test feeder is assumed to
have six switches as shown in Fig. 2. A summary of the
S load ,b 415  j 275 kVA S load ,c 430  j 280 kVA results is given in Fig. 3-4. Only one case when switch 2 open
The minimum voltage corresponding to this load is is given as the typical example which is indicated as sw2 in
Fig. 2. The topology error detection table shows the weighted
Vmin V21, a 0.9402  3.057 sum of squares value in each case. By checking the smallest
This indicates a heavy loading condition on the feeder. value, the final result is also represented in the Topology error
Measurements assumed are also given in the figure: voltage detection table.
and power flow at the substation, current measurements on
branches 6-7, and voltage measurements on node 8, 10 and
17.

Figure 3. Case 1 topology error detection result when switch 2 open

Figure 2. One-line diagram of the test feeder with switch

The topology error processing was implemented using C


language on Microsoft Visual Studio .NET 2003. For testing Figure 4. Case 2 topology error detection result when switch 2 open
topology error detection, all measurements are used and two
different cases were considered : Note that case 1 and case 2 can be used to assess the impact of
Case 1 : Power measurements at the substation (both real and voltage measurements on topology error detection. The results
reactive), indicated as m0 in Fig. 2, plus a current for these two cases indicate that although both cases detect
correct topology error, case 2 which has voltage
measurement on the feeder, m1 , and forecasted load data. measurements smaller than the inequality value which helps to
Case 2 : the same measurements as in Case 1 plus three classify the topology error. Thus, including voltage
voltage measurement m2 ~ m4 from the feeder nodes. measurements helps to detect topology error. In summary, as
these tables indicate, correct topology error detection is
To generate measurement data for testing purposes the
obtained in both cases. Furthermore, in the other switch error,
actual measurements have been obtained by running a power this algorithm detects the topology error correctly.
flow for the given load. Then measurement error was added to
the actual measurements. x Shunt Capacitor Bank Detection
Z
a
Z r eZ 7 For testing shunt capacitor bank (SCB) error detection, the
a same IEEE test feeder is used with the previous test. The test
where Z is actual data and eZ is the measurement error. The feeder is assumed to have two same SCB on nodes 9 and 10 as
forecasted load data is created by perturbing the actual load shown in Fig. 2. SCB is installed to provide capacitive
data by adding error of 30%. The power and current reactive compensation/ power factor correction. Thus, if SCB
magnitude measurement errors are selected from Normal error occurs, it affects reactive power. By using this idea,
4

switch and SCB error is classified. When the inequality value conditions," in Proc. 1993 IEEE PES Winter Meeting., vol. 8,
of reactive power is much larger than that of real power pp. 2009-2015.
[4] Roy Hoffman, "Practical State Estimation for Electric
( fQ  fP ), SCB error occurs. A summary of the results are Distribution Networks," in Proc. 2006 IEEE PSCE Power Syst.
given in Figs. 5&6. Only one case in which SCB 2 open is Conf. and Expo., pp. 510-517.
given as the typical example. Topology error detection table [5] Kersting, W.H., "Radial distribution test feeders," in Proc. 2001
shows weighted sum of squares values in each case. By IEEE PES Winter Meeting., vol. 2, pp. 908-912.
checking the smallest value, the final result is also represented [6] Goswami, S.K. and Basu, S.K., "Direct solution of distribution
systems," IEE Proc. Gener., Transm. and Distrib., vol. 138, pp.
in the Topology error detection table. 78-88, Jan. 1991.
[7] Mesut E. Baran, Jinxiang Zhu, and W. Kelly, Meter placement
for real-time monitoring of distribution feeders IEEE Trans.
Power Syst., vol. 11, pp. 332-337, Feb. 1996.
[8] Mesut E. Baran, Jinxiang Zhu, Hongbo Zhu, and Kenneth E.
Garren, A meter placement method for state estimation IEEE
Trans. Power Syst., vol. 10, pp. 1704-1710, Aug. 1995.
[9] Kevin A. Clements and Antonio S. Costa, Topology Error
Figure 5. Case 1 SCB error detection when SCB 2 open Identification Using Normalized Lagrange Multipliers IEEE
Trans. Power Sys., vol. 13, pp. 347-353, May. 1998.
[10] C.N. Lu, J.H. Teng, and B.S. Chang, Power system network
topology error detection IEE Proc. Gener., Transm. and
Distrib., vol. 141, pp. 623-629, Nov. 1994.
[11] John J. Grainger and William D. Stevenson Jr., Power System
Analysis, McGRAW-HILL, 2008, pp. 641-694.
[12] John G. Webster, Wiley Encyclopedia of Electrical and
Electronics Engineering, John Wiley & Sons, Inc., 1999, pp.
Figure 6. Case 2 SCB error detection when SCB 2 open
137-147.
As these tables indicate, the same result of topology error
detection is obtained. In these cases, Topology error detection VII. BIOGRAPHIES
doesnt work correctly. In this simulation model, the SCB is
very close and has the same capacity. Furthermore, the SCB
Mesut E Baran is currently a professor at North
exists in same meter zone. Therefore, if the SCB which has a Carolina State University in Raleigh, NC. He
similar capacity and is nearby, this algorithm will not detect received his Ph.D. from the University of California,
topology errors correctly. For accurate SCB error detection, an Berkeley in 1988. His research interests include
enhanced algorithm is needed. distribution and transmission system analysis and
control.

V. CONCLUSION
A topology error identification method using BCSE is
presented in this paper. The test results indicate that the
method is capable of topology error detection. This algorithm
Jaesung Jung is currently working towards his
is based on changing the on/off status of branches one after masters in Electrical Engineering at North Carolina
the other and performing a state estimation in each case. If State University in Raleigh, NC. He received his B.S.
after reversing a branch status, residues in one of the in Electrical and Computer Engineering form
ChungNam National University, South Korea in
estimation runs are within the threshold values, then the 2006. His research interests are in the area of power
original branch status is declared false. This algorithm system dynamics and computer simulations in power
successfully detects topology error correctly in every case. systems and power system wide area monitoring and
However, in SCB error detection, When an SCB which has control.
similar capacity and is nearby, this algorithm doesnt work
correctly. For accurate SCB error detection, an enhanced Tom McDermott (SM 1990) is a Senior Consulting Engineer with EnerNex,
algorithm is needed. currently working in wind generation, distribution systems, lightning
protection, custom software development, and electromagnetic transient
studies. He is currently Vice Chairman of the Distribution System Analysis
VI. REFERENCES Subcommittee, a U.S. delegate to IEC TC 57 Working Group 14, and has
[1] Mesut E. Baran and Arthur W. Kelley, "State Estimation for previously chaired the Pittsburgh Section IEEE and the Working Group on
Real-Time Monitoring of Distribution Systems," IEEE Trans. Estimating Lightning Performance of Transmission Lines. Tom is a registered
Power Systems., vol. 9, pp. 1601-1609, Aug. 1994. professional engineer in Pennsylvania. He has a B. S. and M. Eng. in Electric
Power from Rensselaer, and a Ph.D. in Electrical Engineering from Virginia
[2] Mesut E. Baran and Arthur W. Kelley, "A branch-current-based
Tech.
state estimation method for distribution systems," IEEE Trans.
Power Systems., vol. 10, pp. 483-491, Feb. 1995.
[3] Royetelman I. and S.M. Shahidehpour, "State estimation for
electric power distribution systems in quasi real-time

S-ar putea să vă placă și