Validation of a Methodology for Service Restoration on a Real Brazilian Distribution System

Transcription

1 Validation of a Methodology for Service Restoration on a Real Brazilian Distribution System Marcos H. M. Camillo, Marcel E. V. Romero, Rodrigo Z. Fanucchi COPEL Distribuiçao S/A Londrina, Brazil Telma W. de Lima, Leandro T. Marques, A. B. C. Delbem, João B. A. London Jr. University of Sao Paulo Sao Carlos, Brazil Abstract Recently a practical and efficient methodology for service restoration in distribution systems was developed. This methodology combines Multi-objective Evolutionary Algorithms with the tree encoding named Node-Depth Encoding. In comparison with other methodologies already proposed for service restoration, the novel features of this methodology are: (i) to generate adequate service restoration plans for large scale distribution networks (networks modeling distribution systems with thousand of buses and switchers) with relatively soft computing without requiring any network simplification; and (ii) to generate service restoration plans for multiple-faults as good as for a single fault. This paper reports the experience in using that methodology to generate service restoration plans in one real distribution system of COPEL, a company of the Brazilian Electricity sector. More specifically, this paper reports: the main points of that methodology, the analysis of the service restoration plans generated by it performed by engineers of COPEL, and also some suggestions of these engineers in order to improve the methodology. Index Terms Real Distribution System, Service Restoration, Mulitple Faults, Node-Depth Encoding, Multi-Objective Evolutionary Algorithms. I. INTRODUCTION Electric distribution systems are subject to permanent faults. Consequently, the fast service restoration brings customer satisfaction and reduces the compensation paid by power distribution companies. The process of restoring power supply to the healthy out-of-service areas (known as service restoration problem), after the areas with faults have been identified and isolated, requires the use of existing resources and equipment installed in the Radial Distribution Systems (RDS) to seek the best way to restore as many out-of-service loads as possible. This process is much based still in the personal characteristics of each operator of the Distribution Operation Centers (DOC). This is due mainly to the difficulties of viewing all the variables involved in obtaining the solution, the number of possible solutions and the time necessary by human being to find an adequate solution. Based on the above it is extremely important to have computational tools to generate adequate Service Restoration Plans (SRP) in a short time, allowing real time application. An adequate SRP minimizes both the number of out-of-service loads and number of switching operations (the objectives of the service restoration problem) without violating the radiality and operational (limits for the bus voltage, network loading, and substation loading) constraints (the constraints of the service restoration problem). Among the techniques proposed for solving the service restoration problem in large-scale RDSs, there are those based on Multi-Objective Evolutionary Algorithms (MOEA), including the methodology known as Multi-objective Evolutionary Algorithm with Node-Depth Encoding (MEAN) [4]. This methodology combines a technique of MOEA based on subpopulation tables, where each table stores the found SRPs that better attend an objective or a constraint of the problem, with a tree encoding named Node-Depth Encoding (NDE) to computationally represent the RDS electrical topology. As it was demonstrated in [4], NDE can improve the performance of MOEAs to solve the service restoration problem because of the following NDE properties: (i) it has two operators that generate exclusively feasible configurations, i.e., radial configurations able to supply energy for all out-of-service areas having switch linking them to energized areas; (ii) NDE can generate more feasible configurations in comparison to other encodings in the same running time; and (iii) NDE enables a more efficient forward backward Sweep Load Flow Algorithm (SLFA) for RDSs. As a consequence, NDE enables the analysis of electrical aspects (limits for the bus voltage, network loading, and substation loading) of each found SRP in a very fast way. In [1] it is presented the methodology named MEAN with Multiple criteria tables and alarming Heuristic (MEAN-MH). The MEAN-MH is an extension of the MEAN, in order to accommodate new tables, related to the concept of nondominated solutions, and an heuristic to prioritize switching in feeders with larger voltage drop and loading. In order to demonstrate the practical viability of the MEAN-MH, this paper reports the experience of COPEL engineers in using it to obtain adequate SRPs in the events of single and multiple faults in the real RDS of Londrina city, in Parana state, in Brazil (COPEL is a company of the Brazilian Electricity sector). II. SERVICE RESTORATION IN RDS The occurrence of service (energy) interruptions in distribution networks, although undesirable, is a common /14/$ IEEE

2 situation due to various reasons such as expansion works, preventive maintenance interventions or the action of some element of protection due to permanent faults. Distribution networks are generally structured in meshes, but operated in radial configuration for a better coordination of their protective schemes. Therefore service restoration problem usually involves network reconfiguration procedures, that is, the process of altering the topological structure of RDS by opening sectionalizing (normally-closed (NC)) switches and closing tie (normally-open (NO)) switches. However, during the process of service restoration several related issues must also be considered as described: the service must be restored to the maximum possible healthy out-of-service areas; the number of switching operations should be kept to a minimum possible; system radiality must be maintained; and the operation constraints (limits for the bus voltage, network loading, and substation loading) must be respected. Based on the above, the service restoration problem is usually formulated as a multi-objective and multi-constrained optimization problem. A. Representation of RDSs by Graphs The electrical topology of a RDS can be computationally represented by a graph [5]. In this representation, the sectors 1 are represented as nodes of the graph and the switches by the edges interconnecting the sectors. Fig. 1 shows a graph representing a RDS with four feeders (each feeder is represented by a graph tree), where the edges in solid lines represent NC switches and the edges in dashed lines represent NO switches. B. Node-Depth Encoding Fig. 1 RDS with four feeders. NDE is basically a representation of a graph tree in a linear list containing the tree nodes and their depths 2. The list consists of pairs (ni, pi), where n i is the tree node p i is the node depth. Figure 2 shows an example of a graph tree and its NDE. In order to represent a graph forest (a RDS configiration) by NDE it is necessary to use an array of pointers, where each one indexes an NDE of a tree. 1 A sector is a group of buses and lines without switches. 2 The depth a node is the length of the unique path from the root of the tree to the node. C. Operators of the NDE Two operators were developed to efficiently manipulate a forest, encoded by the NDE, producing a new forest, also encoded by the NDE [2]: Preserve Ancestor Operator (PAO) and Change Ancestor Operator (CAO). Each operator modifies the forest encoded by the NDE arrays, which is equivalent to pruning and grafting a subtree of a forest generating a new forest. Both operators are computationally efficient to construct a new forest. Additional information about the NDE and its operators applied to DS reconfiguration problems is provided in [2], [4]. Fig. 2 Tree of a graph and its NDE D. Mathematical formulation of the Service restoration problem The service restoration problem can be formalized as follows [1]: 0 Min. (G) e ψ(g, G ) Subject to: Ax = b; X(G) 1; B(G) 1 e V(G) 1, where: G is a forest of the graph representing a RDS configuration (each tree of the forest corresponds to either a feeder or an out-ofservice area); (G) is the number of customers that are out of service in a configuration G; (G, G 0 ) is the number of switching operations necessary to reach a configuration G from the configuration just after the isolation of the faulted areas G 0 ; A is the incidence matrix of G; x is the vector of line currents in G; b is a vector containing the complex currents in the load buses (with b i 0) or current injections in the buses of substations (with b i 0); X (G) is the highest value of the network loading on the configuration G, given by the biggest x j x j ratio, where x j is an upper bound limiting of current, for each x j line current in a line j; B (G) is the largest value of substation loading in configuration G, given by the biggest ratio b s / bs, where bs is an limiting upper bound for each current injection bs provided by a substation s and V (G) is the higher voltage drop in a configuration G, given by the highest vs v k value, where v k is the voltage magnitude in p.u., at bus k, and v s is the voltage magnitude in p.u., in the substation bus s that feeds the bus k, and is the maximum permissible voltage drop.

3 The formulation of the service restoration problem presented in (1) can be synthesized by considering [1]: (i)penalties for violated constraints X(G), B(G) and V(G); (ii)use of the NDE, i.e. an abstract data type for graphs that can efficiently manipulate a network configuration and guarantee that the modifications always produce a new configuration G that is also a spanning forest (a feasible configuration); (iii)arrangement of the nodes in the Terminal Substation Order (TSO) for each produced configuration G in order to solve Ax=b using an efficient SLFA for RDSs. Through x obtained from a backward sweep, the complex node voltages are calculated from a forward sweep; (iv) (G). The NDE always generates forests that correspond to networks without out-of-service consumers in the reconnectable system. These modeling aspects enable to rewrite (1) as: s.t. 0 Min. ψ(g,g ), X(G), B(G) and Load Flow calculated using NDE, G is a forest generated by the NDE. III. MEAN-MH V(G) The MEAN-MH [1] is based on the concept of subpopulation tables, i.e. each subpopulation table stores the best solutions (or the best RDS configuration) found according to one of the objectives presented in (2) or a function aggregating objectives. Besides these tables, MEAN-MH possesses three additional subpopulation tables that store solutions considering the concept of non-dominated solutions. The idea is to divide a set of M solutions into several fronts (F 1, F 2,...,F k ) according to the degree of dominance of each solution. A solution G i dominates another G j if G i is better than G j according to at least one objective and G i is not worse than G j in all other objectives. The three MEAN-MH subpopulation tables store the solutions of fronts F 1, F 2 and F 3 (F 1 front - called Pareto Front - contains the non-dominated solutions of the whole set M of found solutions, F 2 front contains nondominated solutions of the set M\F 1, F 3 front contains nondominated solutions of the set M\(F 1 F 2 )). The whole set of subpopulation tables of MEAN-MH is: 1. Individuals with low values of an aggregation function defined as: f agg ( x) ( G, G 0 ) ( G) w X X ( G) w B B( G) w V( G), (3) where (G) are the power losses, in p.u., of configuration G, w X, w B and w V are weights balancing among the networks operational constraints. Here, these weights have been set as follows: 1, if, I( G) 1, where I X, B and V wi 0, otherwise. Note that (G) is used in the aggregation function since the previous researches presented in [4],[5] showed it performs V as an interesting weighting between V(G) and X(G), which are very important constraints for service restoration problem (equation (1)); 2. Individuals with low values for (G); 3. Individuals with low values for V(G); 4. Individuals with low values for X(G); 5. Individuals with low values for B(G); 6. Individuals with 1 pair of switching operations beyond 7. Individuals with 2 pairs of switching operations beyond 8. Individuals with 3 pairs of switching operations beyond 9. Individuals with 4 pairs of switching operations beyond 10. Individuals with 5 pairs of switching beyond the initial restoration; 11. Individuals with 6 pairs of switching operations beyond 12. Individuals of Subpopulation with "non-dominated" solutions F 1 ; 13. Individuals of Subpopulation with "non-dominated" solutions F 2 ; 14. Individuals of Subpopulation with "non-dominated" solutions F 3. The size of tables from 1 to 11 is 5 indicating that those tables can store up to 5 solutions. The sizes of Tables 12, 13 and 14 are, respectively, 20, 40 and 40. These values were assigned after performance tests, presented in [1] e [4]. PAO and CAO are the reproduction operators to generate new individuals. Subpopulation table T i receives a new individual I new if T i is not full or if I new is better (according to the criterion associated with Ti) than the worse solution in T i, replacing it. Tables 12 to 14 are related to non-dominance ranking and must be fulfilled according to the dominance ranking. Two criteria are used by MEAN-MH to evaluate dominance: (i) number of switching operations and (ii) aggregation function f agg (G) (3). The MEAN-MH heuristic focuses the application of PAO and CAO (i.e., the opening and closing of switchers) on RDS regions that actually demand reconfigurations, that is, to feeders of relatively high voltage drop and loading. The idea is to restrict the search space and thus obtaining solutions with the lowest number of operations and that do not violate the operational limits of the system. More details about this heuristic can be found in [1]. IV. RESULTS OF TESTS ON A REAL DISTRIBUTION SYSTEM This section presents the results of the application of MEAB-MH to the RDS of Londrina city in operation in This RDS has 30,156 buses, 2,660 NC switches, 250 NO switches and supplies more than 231,000 consumers. Besides, it has 6 substations 138kV/13.8 kv and 64 feeders, totaling a

4 processing capacity of MVA power. To validate the MEAN-MH 50 simulations were performed and analyzed considering single and multiple faults in that RDS. The MEAN-MH was run using a computer with Intel Core i7-3770, 12GB RAM, Linux distribution Ubuntu 12.4 and GCC 4.4 processor with C language compiler. The parameters used to run the MEAN-MH were: 1. Maximum number of individuals generated and evaluated: 70,000; 2. Maximum permissible voltage drop: 10% of the nominal voltage at the substation (13,800V); 3. Maximum permissible loading: 100% of the conductor capacity (defined as the network loading) and the transformer capacity (defined as substation loading); 4. Criterion for stopping the algorithm: having reached the maximum number of individuals evaluated. sector, was performed by closing the SS 1532, which is not a Switch Operable under Load (SOL). This situation is undesirable in view of the need for temporary shutdown of the feeder 54, affecting consumers connected to it to closing the aforementioned switch. It is also worth mentioning that there is the option of feeding the section downstream the SS 3593 through SOL (three-pole Oil-insulated switching with manual operation - TOM 0743). However, the selected individual by MEAN-MH did not relate this maneuver. This situation was denoted as "Deficiency 1" A. Simulations for Single Fault Initially, it was considered to occur a single fault in the sector I of feeder 59, as shown in Figure 3. Table I shows the average and standard deviation values of the best solutions found by MEAN-MH over 50 trials (considering the solutions stored in tables 6 to 14). It is important to highlight that the average values for power losses, voltage drop, running time, network and substation loading of the SRPs obtained indicate that MEAN-MH can be applied for real-time applications. TABLE I SUMMARY OF SIMULATION FOR SINGLE FAULTS Parameter Power losses (kw) Voltage Drop Network Substation Running time (s) Average Standard Deviation Fig. 4 Configuration of the feeder 59 after the isolation of the faulted sector through the application of the switching suggested in the SRPs obtained for the first analysis. Also, it was analyzed the individual of generation 54,452 with low value of voltage drop. The initial configuration is shown in Fig. 5 and the individual obtained in Fig 6. One can easily envision that the switching involving the SS 1992 and Three-pole Gas-insulated switching with Manual operation (TGM) 0853, between feeders 16 and 64, as well as the switching involving the Three-pole Gas-insulated switching with Automatic control (TGA) 0659 and the TGM 4522, between the feeders 45 and 47, are not directly related to the switching for service restoration of healthy sectors of the feeder 59. Fig. 3 Configuration of the feeder 59 before the isolation of the faulted sector (sector I) for the first analysis. Analyzing the initial individual obtained after the isolation of the faulted sector, as shown in Figure 4, it was noted that the switching suggested to perform the reconfiguration of the healthy sectors meets the criteria of system loading and voltage drop for the feeders involved with this contingency situation (feeders 59, 32 and 54). However, it was also observed that the recovery of the sector downstream of the single pole disconnect Switch 3593 (SS 3593), i.e. the K Fig. 5 Configuration of the feeders involved with fault before the isolation of the faulted sector (sector I) for the second analysis.

5 TABLE II SUMMARY OF SIMULATIONS FOR MULTIPLE FAULTS Parameter Power losses (kw) Voltage Drop Network Substation Running time (s) Average Standard Deviation Fig. 6 Configuration of the feeders involved with fault after the isolation of the faulted sector through the application of the switching suggested in the SRPs obtained for the second analysis. Looking at the situation it was found that the MEAN-MH uses the voltage drop and the maximum network loading as global variables. In the case under study, analyzing the moment previous top the fault, the feeder with highest voltage drop is the feeder 45, with 2.48%. This feeder has no interconnection with others involved in the restoration of the healthy sectors. Thus, unless the initial configuration results in a higher drop to the displayed value of 2.48%, informed as "outage" it will always be the value of 2.48%. Deepening data on 54,452 individuals, it could be seen that the voltage drop in the initial generation was 3.26%, thus presenting a feasible solution. Thus, switching involving the restoration of the initial level of voltage drop (opening of TGA 0659, TGM 0853 and SS 037 and closing of TGM 4522, SS 1992 and TOM 0236) would probably not be made by DOC, only switching related to rebuilding of healthy sections and isolation of the faulty section (Opening SS 3186, SS 0905 and SS 3593 switches and closing the TOM 0234 and TOM 0743 switches) would be conducted. This situation was defined as Deficiency 2. B. Simulations for Multiple Faults To check for multiple faults in RDS Londrina, the system in question was simulated considering the fault in 5 sectors belonging to different feeders, all from the same bus of a given substation. Thus, the simulated impact was the total loss of one of the buses of the substation in need of load transfer using only network equipment. Table II summarizes, considering the best individuals present in tables of pairs of switching and "non-dominated" solutions of all 50 simulations. Through the analysis of individuals generated in the different simulations, we noticed the same "Deficiency 2" presented for all the cases of simple faults. As an example, in 45,272-individual of the simulation 1 there is a transfer of loads between feeders 42 and 49, and between 40 and 42. These transfers reduce the total resistive losses of the RDS; however, as already shown, it has no direct relationship with the service restoration process in question. It should be noted however that these situations were less frequent compared with those occurring in cases of simple faults. This is due to the concentration of faulted sectors in a small geographic region sectors. Thus, the points with larger loads and voltage drops of the overall system became more restricted to the problem area (favoring the selection of feeders with direct interconnection with those missing). Finally, analyzing the starting individual of simulation 25 present in the non-dominated solutions table, there was the proposition of operating the SS 1532, SS 5344 and SS 150. For the same section we have that load transfer could have been made to the switch of the TGA 3815, TGA 376 and TOM 742 (all switchable under load). Thus, we conclude that the same "1 Deficiency" shown in cases of simple faults also occurs for the case of multiple faults. V. DISCUSSION AND CONCLUSION This paper evaluated the methodology for service restoration called MEAN-MH, which is based on multiobjective EAs and at the NDE. This assessment occurred through the application of MEAN-MH to obtain SRPs after the occurrence of permanent faults (simple and multiple) in the COPEL RDS of the Londrina city. In general the MEAN-MH showed quite satisfactory results and consistent with those present in [1] and [4]. By the application in a real and large-scale RDS and with analysis of specialists in the area, it was still possible to validate the consistency of the results obtained by MEAN-MH, whose final configurations showed feasible solutions close to that expected for the area (DOC). For the case of simple faults, the existence of some deficiencies was clearly observed, whose analysis allowed proposing improvements for the MEAN-MH, improvements that raise the increasingly closer to the result expected by the electricity dealerships. As for the case of multiple faults, the same deficiencies were observed, but with a lower incidence when compared with the case of simple faults. In view of engineers of COPEL, among the proposals to be presented below, the numbers 1, 2 and 5 have a high level of applicability for single and multiple faults. Already the number of 3 and 4 have greater impact in cases of simple faults.

6 Proposals for MEAN-MH improvement: Regarding the "Deficiency 1": Proposition 1: Review all possible initial configurations for the generation of the initial individual, in order to increase the likelihood of identifying individuals (solutions) with reduced amount of switching; Proposition 2: Consider the existence of different types of switchgear (existence of remote control, possibility of opening with load and / or short circuit, etc). Thus, one sees increased speed of operations (using, for example, switches with remote control and command) and reduction in the quantity of customers affected (by using operable switches to prevent momentary load opening feeders for switching). Regarding the "Deficiency 2": Proposition 3: change in philosophy of recording the value of the maximum voltage drop and maximum loads. The new philosophy would consider only values of feeders that are involved in the restoring power maneuver, i.e. those with abnormal initial state of their switches to maneuver; Proposition 4: application of a new heuristic in which the prioritization of feeders to be chosen for switching occurs according to the electrical connectivity of these circuits with the missing power supply. Propositions 3 and 4 seek to limit the search space to the region next the missing sector, avoiding the generation of switching in feeders that do not contribute to the process of service restoration. During the simulations it was found that the value displayed for maximum network loading took into account only the limit od allowable current of conductors and switches (elements currently available on the MEAN-MH database). However, in practice, the maximum load of the network is also defined by the current limit of the connections between conductors. Thus, it is also proposed: Proposition 5: use the limit values of the existing connectors in distribution networks, besides the limits of conductors and switches to define the threshold current values from power conductors. Given the above, it is expected that the reported study and the improvements suggested in this paper can help the development of new methodologies to solve the service restoration problem. And, as consequence, to turn more reliable RDSs around the world, in order to bring satisfaction for the customers and to reduce the time out of service in contingency situations. ACKNOWLEDGMENT This paper is part of a project of Research and Development of Brazilian Electricity Regulatory Agency (PD /2012). The project is ongoing and includes the participation of members of the USP, UFG and COPEL. In this project, the improvements suggested in this paper will be studied and considered to the development of a practical methodology for service restoration. REFERENCES Periodicals: [1] D. S. Sanches, A.C.B. Delbem, J. B. A. London Jr, Multi- Objective Evolutionary Algorithm for single and multiple fault service restoration in large-scale distribution systems. Electric Power Systems Research (Print), [2] A. C. B. Delbem, A. C. P. L. F. de Carvalho, C. A. Policastro, A. K. O. Pinto, K. Honda, and A. C. Garcia, Node-depth encoding for evolutionary algorithms applied to network design. GECCO, pp , [3] K. Deb, S. Agrawal, A. Pratab and T. Meyarivan, A fast and elitist multiobjective genetic algorithm: nsga-ii. Evolutionary Computation, IEEE Transactions on, [S.l.], v.6, n.2, p , Apr [4] A. C. Santos, A. C. B. Delbem, J. B. A. London, Jr., N. G. Bretas, "Node-Depth Encoding and Multiobjective Evolutionary Algorithm Applied to Large-Scale Distribution System Reconfiguration", IEEE Transactions on Power Systems, Vol. 25, No. 3, August [5] A. C. B. Delbem, A. C. P. L. F. Carvalho, N. G. Bretas, Main Chain Representation for Evolutionary Algorithm Applied to Distribution System Reconfiguration. IEEE Transactions on Power Systems, v. 20, n. 1, p , 2005.