Distribution Feeder Reconfiguration For Operation Cost Reduction

IEEE Transactions on Power Systems, Vol. 12, No. 2, May 1997

730

Distribution Feeder Reconfiguration For Operation Cost Qin Zhou Consultant San Francisco, California

Dariush Shirmohammadi Shir Consultants San Ramon, California

Abstract - This paper describes a new feeder reconfiguration algorithm for the purpose of reducing the operating cost in the real-time operation environment. The methodology developed is a heuristic based approach. It emphasizes on minimizing the cost of operation over a specified time period rather than a fixed operating point. The practical operating concerns of feeder reconfiguration and the coordination with other distribution automation applications are also addressed. The developed algorithm has been implemented as a production grade software. Test results on PG&E distribution feeders show that the performance of this software is efficient and robust.

Keywords: distribution automation, feeder reconfiguration, cost reduction, optimization, heuristic rules. I.

INTRODUCTION

Feeder reconfiguration entails altering the topological structure of distribution feeders by changing the opedclose status of the switches in these feeders under both normal and abnormal operating conditions. The benefits of feeder reconfiguration include restoring power to outaged portion(s) of a feeder in a timely manner whch improves the value of service to customers by reducing average outage time, relieving of overloads on feeders by s h f i n g load in real-time to adjacent feeders which could defer capital expansion projects, and reducing resistive line losses which could reduce the operating cost of a distribution system. In this paper we present a reconfiguration algorithm for reducing the operating cost of distribution systems. In recent years, considerable research has been conducted in the area of loss minimization [l]. Most of these algorithms are developed for reducing the feeder resistive losses and they are basically planning tools. The following important issues have not been fully addressed in the previously published literature: 96 SM 512-4 PWRS A paper recommended and approved by the IEEE Power System Engineering Committee of the IEEE Power Engineering Society for presentation at the 1996 IEEHPES Summer Meeting, July 28 - August 1, 1996, 1996, in Denver, Colorado. Manuscript submitted January 2, 1996; made available for printing May 17, 1996.

0 0

0

0 0

W.-H. Edwin Liu Pacific Gas and Electric Company San Francisco, California

Application to real-time operation environment. Consideration of all operating constraints, such as limit on the number of switching operations. Reduction of operating cost over a specified time period, which includes ensuring that a specific set of switching operations that reduce losses at a fixed operating point will not raise losses at a later time. Consideration of the cost of switching operations and the sequence of implementing switching operations. Consideration of protection requirements. Coordination with other distribution automation functions.

The algorithm presented in this paper focuses on providing an operation decision support tool. It focuses on minimizing the cost of operation over a specified time period rather than simply reducing losses for a fixed operating point. The practical operating concerns of feeder reconfiguration and the coordination with other distribution automation applications are also addressed in the paper. We developed a production grade software, named FlRECON, based on the algorithm presented in this paper and have implemented it for use with PG&E distribution systems. In the following we first present a brief description of the state-of-the-art feeder reconfiguration techniques for loss reduction.

II. STATE-OF-THE-ART TECHNIQUES Feeder reconfiguration for loss reduction entails finding a series of opedclose switchmg operations to reduce the resistive losses in primary distribution feeders. There have been many papers written in this area. They can be classified into the following categories: 0 Composite heuristics and optimization methods: A well known method in this category was proposed by Merlin and Back [2] and later modified by uses branch-andShirmohammadi and Hong_[3].. It ~ bound optimization technique along with practical heuristics to find near optimum configuration for a distribution system. Heuristic based methods: The heuristic based attempt to find the optimum switching operations one pair at a time. Each pair of opedclose switching operations reduces certain amount of resistive losses. These methods use empirical formulas to assess the loss reduction associated with each switching operation and introduce the rules to reduce the number of

.

0885-8950/97/$10.00 0 1996 IEEE

73 1

candidate switching operations. A good example is the algorithm developed by Civanlar and Grainger [4]. They use two well founded heuristic rules to select each switching operation; 1) loss reduction can only be attained if there is a s i m i c a n t voltage difference across an open switch, 2) loss reduction will be achieved if loads on the higher voltage drop side of the open switch are transferred to the other side. An empirical formula is used to evaluate the loss change due to a pair of switching operations without running a k l l power flow. Thus, the search process is very fast and efficient. ,4rtificial intelligence (AI) based methods: In recent years the AI techniques such as artificial neural inetworks (ANN), genetic algorithms (GA) and expert systems have been used to determine the distribution system reconfiguration [5], [6], and [7]. The ANN is a potential candidate for on-line applications because of iits fast computational performance. The major drawback is that it requires substantial amount of accurate data for training and the training has to be (conducted for each distribution network and the ;subsequent changes in the network must be accounted €or, The GA methods can provide the solution which is independent from the initial system configuration lbut the computation speed is too slow to apply to large !distribution systems. Expert system based approaches 'basically work similar to the heuristic based methods. Although there have been extensive research work on the issue: of loss minimization, it is still a rare practice to operate switches in real-time purely for reducing network resistive losses. This is because many important real-time operation considerations, described earlier, are not fully addressed by any of the family of methods described above. In order to account for all considerations necessary for realtime operating environment, we made the following choices in thLe development of cost reduction algorithm: Only consider the operation of remote-controlled switches. Consider all the relevant operating constraints such as limit on the number of switching operations. Aim for minimizing the cost of system operation over a time period rather than minimizing resistive losses for a fixed operating point. 0 Verm the solution by a real-time relay coordination program to ensure proper protection for the new configuration.

III. METHODOLOGY Cost reduction due to an operdclose switching operation for the time period T can be calculated as,

AC

= -jc"(t)AP(t)dt-(CSo 0

+CS,)

(1)

AC :

cost change due to an opedclose switching operation ($) ckwh ( t ): cost of energy ($/kwh) for time t T: time period defined by the operator during which no further switching operations will be carried out (hour) AP( t ): loss change (kw) for time t CSo: operating cost of closing an open switch ($) CS, : operating cost of opening a closed switch ($) The discrete version of (1) is: N

AC = - 2 C j k W h h p j - (CS, + CS,) j=1

(2)

Herej means the j t h time interval. According to reference [4] the loss change, A P ( t ) , due to an opedclose switching operation can be estimated as,

D Rloop

4

m n E, E, w.1,

set of nodes that are disconnected from one feeder and connected to another feeder sum of the resistance in the path of the loop formed when the normally open switch is closed nodal current injection of ith node starting node number of normally open switch end node number of normally open switch component of resistive node voltage drop at node m component of resistive node voltage drop at node n

*, 1.1 real part, complex conjugate, and magnitude operators, respectively

The resistive node voltage drop is obtained as, E=RI Where, E is the vector of resistive node voltage drops I is the vector of nodal current injections R is the real part of node impedance matrix

(4)

Thus, the loss change at any time intemal j, A P j , will be,

(5) where AT is the length of time interval. By substituting ( 5 ) into (2) we can calculate the cost change due to an opedclose switching operation for the time period T. Overall Cost Reduction Algorithm: The framework of cost reduction module consists of the following steps. In a distribution feeder system consisting many normally closed and normally open switches: 1. Check the limit on the number of switching operations for each switch. If the number of switching operations of switch k carried out in the last H hours (where H is specified by the operator) has reached the limit of

~

732

switching operations for that switch discard this switch from further consideration. 2. Perform some topology processing: 0 I d e n m normally closed switches "correlated" to each normally open switch (here "correlated" means that those closed switches are in the loop(s) formed by closing a normally open switch). 0 Determine the capacity margin of each normally open switch. 3. Select one candidate opedclose switching operation. The selection of opedclose switching operation involves four steps: a) Identlfy normally open switch candidates for closing: 0 Divide the time period T into N equivalent intervals AT and run power flow for each interval based on the load forecast information. 0 For each normally open switch k and each time interval j Q=1,2, ..., N), calculate resistive node voltage drops EA and E,' based on (4). 0 Sum resistive node voltage drops of open switch k over these N intervals as follows:

If ACm"r I 0, ignore this open switch k Repeat above steps for all normally open switches to narrow down the selection of normally open switch candidates. All remaining normally open switches are candidates for closing. If all normally open switches are eliminated at this stage, then go to step 7. For each open switch candidate, eliminate undesirable closed switches from consideration: For each normally open switch candidate k identified in step 3a:

then ignore the closed switches on the nth node side of switch k.

then ignore the closed switches on the mth node side of switch k. (See Appendix B for heuristic rules applied for the above elimination.) The remaining closed switches will be the candidates for opening when we close normally open switch k. Identlfy and rank all remaining opedclose options in reducing value of AC: For all available opedclose switching options consisting of the combinations of all candidates of open switches identifed in 3a and their correlated candidates oi closed switches identlfied in 3b: Calculate loss reduction for each time interval via (5); Calculate cost reduction AC over the entire time period T via (2); Rank all open/close options in reducing value of AC. If all AC values of these opedclose switching options are negative, then there is no more cost reduction and go to step 7. Select the opedclose switching option with maximum cost reduction: Select the opedclose switching option with maximum cost reduction from the list identlfied in 3c:

then ignore this normally open switch k from further consideration. E, is the tolerance defined by operator. (See Appendix B for heuristic rules applied for the above elimination.) Evaluate maximum possible cost saving for each remaining normally open switch candidate: The approximate maximum loss reduction corresponding to switch k in time interval j could be derived from (5) whch is (see Appendix A for derivation):

"loop

The minimum switching cost of open switch k and its correlated closed switch is: CSo, + MinCSc (7) CSek is the switching cost of the normally open switch k and MinCSc is the minimum switching cost of those closed switches which are in the loop formed by closing the normally open switch k. Substituting (6) and (7) into (2), the maximum cost reduction corresponding to this normally open switch k is: AC"k

N

= -CcjhhAPjm"k

i

-

(CSOt+ hfincs,)

(8)

AC"" = max(AC,,AC?, ..., AC,, ...) (9) Check for capacity constraints violations: For the selected opedclose switching option, if the total load being transferred is bigger than the capacity margin of the selected open switch, then discard this switching option and select the switching option with next highest cost reduction. If no switch option is available, go to step 7. Run power flow for new system configuration tocheck for operating constraints violation for every time 0

4.

.

~

733

interval. If operating constraints violations exist, go to step 6. 5 . Check the limit on the total number of switching operations. Mathematically they can be shown as,

CLrLm?

cost reduction

24.89 !$

Selected opedclose operation No. 2: loss reduction 110.57 kwh 6.85 $ cost reduction

k

C L the total number of switching operations. k

Lm the limit of the overall number of switching

operations. If the answer is yes, then no more switching operation is allowed, go to step 7; otherwise go to step 2. 6. Discard the selected switching option and go to step 3d to pick up the opedclose switching option corresponding to the next highest cost reduction. 7. Optimal solution achieved. The total cost reduction of next T hours after completing all switching operations is: N ACtotal = - C C J h h P J

-

J=1

csk

k EM

(10)

Where,

M: CS, : :

Set of switches whose status should be altered cost of operating switch k loss reduction for time interval j (it is the

difference between the resistive losses of new system configuration after completing all switching operations and the losses of original system configuration. It is calculated by running power flow for each time interval j.) T, N and cihh are defined as before.

N.NUMERICAL RESULTS The cost reduction algorithm presented here was used to develop production grade FRECON program which was successfully tested on many PG&E’s distribution systems. The following is only one example of the test results. The test system consists of two 12kv feeders with 920 line sec1:ions and 55 switches, which can be operated for the cost reduction purpose, with peak load of 4.85MVA over a 24 houw period. The cost reduction algorithm picks up the opedclose switching pairs one by one based on the heuristic rules described in section 111. The limit of overall number of switching operations is 6 over the 24 hour time period. The cost reduction values were evaluated based on the assumption that the cost of energy is $0.08 per kilowatt hour and the cost of performing switching operation is $1.00 per switching operation (remember that only remotecontrolled switches are considered here). The test results of running the cost reduction program for this 2 feeder systems are shown as follows: Selected opedclose operation No. 1: loss reduction 336.18 kwh

Selected opedclose operation No. 3 : No opcdclose option can reduce the system operation cost ! ! Final optimal solution: total loss reduction total cost reduction

: :

446.75 kwh 3 1.74 $

The test results show that, Cost reduction can be achieved by two pairs of switching operations and no more switching pair can reduce the cost any further. 0 The total loss reduction is 446.75 kwh after conducting the switching operations recommended by the cost reduction program. The total loss reduction is 8.66% which is significant. 0 The cost reduction value for this test system is only $31.74 for the 24 hours period. However, the size of test feeders are relatively small. Considering that there are around three thousand feeders in PG&E distribution systems, the total cost reduction by conducting the feeder reconfiguration could be significant. 0 The developed algorithm is computationally efficient. For the test system above, it took less than 15 seconds on a SUN SPARClO workstation to find the solution. This elapsed time includes the I/O time, solution search and repeatedly running power flow for all the 24 hours after implementing each pair of switching operations.

V. IMPLEMENTATION ISSUES Sequence of Implementing the Switching Operations After cost reduction program determines the optimal solution, which is a set of opedclose switching operations, it is necessary to find a sequence for implementing the switching operations. This is necessary since at any intermediate step of switching implementation we must guarantee that there are no major operating constraints violations. The answer is definitely not trivial. We have developed an algorithm to deal with this problem. The objective of this algorithm is to find the sequence of switching operations for implementing the optimal solution of cost reduction program. The principles we used for developing this algorithm are: 0 Always close an open switch first then open a closed switch to ensure continued service during the implementation of switching operations.

~

734 0

0

The open switch that is to be closed next should create the lowest possible loop current (should have the lowest voltage difference across its terminals). This will ensure minimum disturbance during the switching operations sequence. Since no temporary outages are allowed during the reconfiguration, the final sequence of switchmg operations may be a subset of operations determined in the original optimal solution. Also the cost of operating bypass switches, if there are any, must be accounted for. Thus, the final cost reduction value after implementing the sequence of switching operations may be less than the expected one from the original optimal solution.

The following are the major steps of the algorithm for implementing the switching operations: s1. Input the selected switching operations as determined by the cost reduction algorithm. s2. Select an openklose switch pair based on the following logic; 0 The switch to be closed should result in lowest possible loop current; 0 The switch to be opened should create no operating constraints violations. s3. Simulate the switch closing for the selected switch pair and check for constraints violations. There are two kinds of constraints that should be specifically checked, 0 Overloads in the loop formed by the switch closing; 0 Load break capability of the switch that should be opened. s4. Identify the bypass switch candidates if constraints violations exist in step S3. s5. Select the bypass switch if selected closed switch cannot be opened due to the lack of load break capability. The selected bypass switch must be able to 0 relieve the constraints violations generated in step s3; 0 ensure no new constraints violations are generated due to the switching operations of the bypass switch. S6. Go to step S2 to select next switch pair or finish the sequence search. The sequence search will be terminated if 0 all selected switching operations as determined by the cost reduction algorithm have been accounted for, or; 0 no switching operations can be further considered because of the constraints violations.

The following are the major steps in the coordination of cost reduction program with relay coordination program [lo]: P1. Send the following data to relay coordination program: 0 List of recommended switchmg operations. 0 Power flow outputs of the recommended new system configuration. P2. Request from the relay coordination program whether the recommended switching operations are acceptable from protection point of view: 0 If the recommended switching operations are acceptable, the cost reduction program has achieved the final solution and it shall pass the results to operator for his action. If the recommended switching operations are not acceptable, the cost reduction program should discard the selected solution and start a new solution iteration at the next time interval.

VL CONCLUSIONS A new feeder reconfiguration algorithm for cost reduction has been developed and tested as part of the distribution automation systems in PG&E. It provides an operating decision support tool for the real-time operation environment but could also be used as a planning tool. Its function is to find a series of opedclose switching operations to reduce the resistive losses in primary distribution feeders and hence reduce the cost of system operation. The methodology used for cost reduction is heuristic. The algorithm emphasizes timely finding of the solution with as small a number of switching operations as possible. Most practical operating concerns for changing the configuration of the feeder as well as coordination with other distribution automation applications such as relay coordination are also addressed. The test results show that the performance of the developed algorithm is efficient and robust.

VU. ACKNOWLE The authors would like to thank Dr. Carol Cheng and Mr. Ken Lau of PG&E for their useful comments in the development and implementation of t h s software package. PG&E’s R&D department funded the development of the proposed algorithms and the FRECON software program.

VIE.

NCES

R.J. S d i , M.M.A. Salama, A.Y. Chakani, “A survey of the state of the art in distribution system reconfiguration for system loss reduction,” Electric Power Systems Research 3 1 (1994), pp. 61-70.

Protection Considerations From system protection point of view the viability of new network configuration recommended by cost reduction program must be verified to assure that the new configuration can be fully protected by the relay devices.

A. Merlin, H. Back, “Search for a Minimal-Loss Operating Tree Configuration for an Urban Power Dishbution System,” Proc. ofPSCC. Cambridge 1975, Paper 1.2/6. D. Shirmohammadi, H. Wayne Hong, “Reconfiguration of electric distribution networks for resistive losses reduction,”

735 IEEE Transaction on Power Delivery, April 1989, pp. 14021498. [4] S. Civanlar, J.J. Grainger, H. Yin, S.S. Lee, “Distribution feeder reconfiguration for loss reduction,” IEEE Trans. Power Delivery, 3 (1988), pp. 1217-1223. [5] K. Kim, Y. KO and K. H. Hung, “Artificial neural network based feeder reconfiguration for loss reduction in distribution systems,” IEEE Trans. Power Delivery, 8 (1993) 1356-1366. [6] K. Nara, T. Satoh and M. Kitagawa, “Distribution system loss minimum re-configuration by genetic algorithm,” proc. 3rd Symp. Expert systems application to Power System (ESAPS).Tokyo and Kobe, Japan, 1991, pp. 724-730. [7] T. Taylor and D. Lubkeman, “Implementation of heuristic search strategies for distribution feeder reconfiguration,” IEEE Trans. Power Delivery, 5 (1990) 239-246. [SI Mesut E. Baran, F. F. Wu, ‘Wetwork reconfguralion in distribution systems for loss reduction and load balancing,” IEEE Trans. Power Delivery, 4 (1989) 1401-1407. [9] J.C. Wang, H. D. Chiang, G. R. Darling, “ An efficient algorithm for real-time network reconfiguration in large scale unbalanced distribution systems,” PICA 1995, pp. 510-516. [lo] FSoudi and J. Yee,“How distribution automation and protection systems can complement each other,” proceeding of the 21st annual western protective relay conference, Spokane, WA, Oct. 1994. Qin Zhou (M’92) received his B.S. and M.S. in Electrical Engineering from Tsinghua University in Beijing, China in 1983

and 1986 respectively, his Ph.D. in Electrical Engineering from the Iowa State University in 1992. Between 1986 and 1989, he work.ed as a lecturer in the Dept. of Electrical Engineering, Tsinjghua University. Since 1992 he works as a consultant at Pacilic Gas and Electric Company (PG&E). Qin Zhou is a member of the Tau Beta Pi Honor Society. Dariush Shirmohammadi (SM89) is the principal consultant

and founder of Shir Consultants. He received his B.Sc. in Eleciincal Engineering from Sharif University of Technology in 1975’and M.A.Sc. and Ph.D. in Electrical Engineering from the University of Toronto in 1978 and 1982 respectively. Between 1977‘and 1979, he worked in Hydro Quebec Institute of Research (IREQ) on the subject of external insulation. Between 1982 and 1985, he worked in Ontario Hydro on the development and implementation of the EMTP. Between 1985 and 1995 he worked in the Pacific Gas and Electric Company (PG&E) where his last position was the Director of Energy Systems Automation group responsible for developing and implementing distribution automation technologies. Dariush is a registered Professional Engineer in the Province of Ontario, Canada. Wemi-hsiung Edwin Liu (SM’94) received his B.S. degree from National Taiwan University in 1981. He received his M.S. degree in 1’384 and Ph.D. degree in 1987 both from the University of California, Berkeley, in Electrical Engineering and Computer Scieinces. From 1987 to 1991, he worked for Empros Systems International. He has been with PG&E since July 1991 where he is responsible for several R&D projects. Dr. Liu also taught graduate courses in San Francisco State University.

Appendix A

Derivation of Equation (6)

In equation (9,assume E l i is a continuous real variable, and itD

E n ,Em are real numbers, we have,

Take the derivative of

AP with respect to Ii and let it equal to ED

zero, then

E, _ -E._ XIi = -_ Rim, Substitute (A2)into (Al), the maximum loss change of AP is: i d

(A31 Rk“ which is equation (6). Appendix B According to reference [3], loss change, h P ( t ) , due to an opedclose switching operation can be calculated as,

Based on this equation, heuristic rules may be developed for selecting most suitable switching operations. For example, two heuristic rules can be applied for identifyingnormally open switch candidates and eliminating undesired closed switches from consideration for each open switch candidate. These two heuristic rules are: If for an open switch candidate E, and En are close to each other at all time intervals in the time period T, then any opedclose switching operation associated with this open switch may not result in a significant loss reduction due to the domination of the second item of equation (Bl). Thus, we can remove this normally open switch from further considerations. The loss reduction can be obtained only by transferring loads from the side of the normally open switch with higher resistive node voltage drop to the side of the switch with lower resistive node voltage drop because only in this case the change in losses could be negative. Thus, we do not need to consider those closed switches on the side of open switch with lower resistive node voltage drop.