A Deterministic Analysis Method for Back-feed Power Restoration of Distribution Networks

Transcription

1 1 A Deterministic Analysis Method for Back-feed Power Restoration of Distribution Networks Zhenyuan Wang, Vaibhav Donde, Fang Yang, James Stoupis, Member, IEEE Abstract This paper presents a deterministic algorithm that identifies a back-feed restoration strategy to restore the out-ofservice load due to fault isolation while ensuring that the postrestoration network has a valid configuration. The algorithm is based on the concepts of network tracing and it supports both single-path and multi-path restoration. In case the network components are too stressed and even the multi-path restoration cannot restore all the out-of-service loads, the algorithm tries to shed minimal load while restoring as many other loads as possible. Capability of the algorithm is demonstrated with a few back-feed restoration solution examples. Index Terms feeder automation, distribution automation, restoration switching analysis, power restoration, network model. S I. INTRODUCTION MART Grid refers to electric power systems that enhance grid reliability and efficiency by automatically anticipating and responding to system disturbances [1]. To achieve smart grid at the power distribution system level, various automatic technologies have been attempted in the areas of system metering, protection, and control. Within these technologies, automated power restoration [2-4] is an important part of the smart grid puzzle. Traditionally, electric utilities use the trouble call system to detect power outages. Specifically, when a fault occurs and customers experience power outages, they call and report the power outage. The distribution system control center then dispatches a maintenance crew to the field. The crew first finds the actual fault location, and then implements the switching scheme(s) to conduct fault isolation and power restoration. This traditional procedure for power restoration may take several hours to complete, depending on how fast customers report the power outage and the maintenance crew can locate the fault point and conduct the power restoration. In recent years, utilities have deployed the feeder switching devices (reclosers, circuit breakers, and so on) with intelligent electronic devices (IEDs) for protection and control applications. The automated capabilities of IEDs, such as measurement, monitoring, control, and communication This work is supported by ABB Corporate Research funding. All authors are with the ABB US Corporate Research Center, 940 Main Campus Dr., Suite 300, Raleigh, NC 27606, USA ( s: Zhenyuan.Wang@us.abb.com, Vaibhav.D.Donde@us.abb.com, Fang.Yang@us.abb.com, James.Stoupis@us.abb.com ) /09/$ IEEE functions, make it practical to implement automated fault identification, fault isolation, and power restoration. As a result, the power outage duration and the system reliability can be improved significantly. Based on the information provided by IEDs, automated fault location identification and fault isolation are relatively easy to achieve. In contrast, automated power restoration becomes a challenging task because of the consideration of operating constraints, load balancing and many other practical concerns, and many research efforts have been focused in this area. Many automated power restoration algorithms have been proposed in previous literature, such as heuristic search-based techniques [5-8], artificial intelligent-based algorithms [9-14] (e.g., expert system, genetic method, fuzzy logic), analyticalbased algorithms [15, 16], and algorithms combining two or more of these techniques [16, 17]. Although some of the proposed algorithms aim to provide a real-time solution, most of them are only suitable for planning analysis or were developed to be executed in the distribution control centers to aid system operators with appropriate decisions. This paper presents an on-line method for the automated power restoration application previously described. The developed method conducts a deterministic analysis to achieve back-feed power restoration, i.e., healthy load zones that have lost power will be restored through their boundary tie switching devices from neighboring sources, and no reconfiguration beyond the tie devices will be considered. The back-feed restoration should not overload any part of the back-feeding network. As forward-feed power restoration for circuit breaker and load switch mixed networks is straightforward, it is excluded from the scope of this method. Thus, the fault isolation switching devices in this algorithm are used for back-feeding isolation only. II. REQUIREMENTS, CONCEPTS AND METHODOLOGIES A restoration switching analysis (RSA) algorithm produces a switching sequence that when executed, will reach a valid post-restoration network that satisfies the following requirements: 1) it is radial; 2) there is no current violation at any network component; 3) there is no voltage violation at any network node. A complete restoration switching analysis algorithm also has optimization requirements such as the minimization of

2 2 network losses and the number of switching operation, and the load balancing of transformers that participate in the backfeed, etc. Not all these requirements are covered in this paper, Furthermore, other important issues related to the fault restoration execution, such as protection coordination of IEDs after the restoration, and switching control exception (switching device fails to operate) handling during the execution, are not discussed in this paper due to the on-going research nature of the topic. They are certainly the subjects of future papers. The algorithm in this paper is based on the following concepts and methodologies. A. Network Model For the sake of method description, a simplified network model is considered that includes three types of components: sources, switching devices (a.k.a. switches that represent sectionalizers, load switches, circuit breakers and reclosers), and loads. Sources are assumed to have limited capacity (ampere rating) but constant voltage. Switches are assumed to have limited loading capability (in amperes, circuit breakers and reclosers have unlimited current interruption capability). Loads are assumed to be of constant power and connect to switches over zero-impedance feeder conductors. The conductors have limited current carrying capability. B. Connectivity Matrix The connectivity of the network model is represented by the branch-node incidence matrix of the graph represented by the network. For the simplified network model in Figure 1, the matrix has the form Brk1 SW 2 SW 6 SW 3 Brk2 SW1 SW5 SW 4 L1 L2 L3 L4 L5 L6 S1 S2 The rows and columns of this matrix correspond to the switches and buses (loads and sources) respectively. The rows and columns are labeled with the corresponding switch and bus names for clarity. Columns of this matrix are arranged in such a way that the ones corresponding to load buses are placed first, followed by those corresponding to source buses. To observe this, the matrix above has been partitioned into two groups of columns. The first group of columns correspond to load buses while the other group to source buses. The upstream and downstream bus information for each (1) switch is also stored in this matrix. The elements on any row are either +1, -1 or 0 (0 is omitted in Equation 1 for clarity). The element +1 denotes the upstream bus position and -1 denotes the downstream bus position. For instance, the first row that corresponds to the switch Brk1, the element in the L1 column is -1 and that in the S1 column is +1. Thus, bus S1 is an upstream bus of Brk1 while L1 is its downstream bus. Since an upstream and downstream of a normally open (NO) tie switch are not defined, +1 and -1 elements are arbitrarily placed at the columns corresponding to the terminal buses of an NO tie switch (for example, see the rows corresponding to switches SW5 and SW6). The network tracing algorithms described below are based on this connectivity matrix. Figure 1: Example Distribution Network (Simplified) C. Restoration Validation Check The restoration validation check confirms the validity of the post-restoration network configuration in order to ensure that the network is radial and all the currents and voltages are within the component limits. The restoration algorithm that forms the main focus of this paper (Section III. ) produces radial post-restoration networks. Thus any additional radiality checks are not necessary. A current violation check is done as an integral part of the algorithm, based on the loading aggregation method described below. This check ensures that for all the network components, their post-restoration loading currents are less than their loading current limits. Voltage violation can be checked after a load flow analysis of the post-restoration network, or it can be checked as an integral part of the algorithm. This paper uses a simplified network model mainly to illustrate how the restoration switching sequence can be determined. For such a simplified network, load flow analysis cannot guarantee the outcome accuracy of the voltage violation check (In fact, any load flow analysis without considering load fluctuations has limited capability of evaluating bus voltage violations, and the voltage violation check based on this analysis is of limited value). Voltage violation check is therefore not the main focus of this paper. However, when a detailed network model is available, this algorithm can be modified to include a voltage validation process based on load flow analysis, to make sure the postrestoration network does not have dangerously high or low voltages on any node.

3 3 D. Network Tracing based Loading Aggregation As stated in the introduction, back-feed power restoration should not overload any part of the back-feeding network. In the algorithm, this is achieved by a recursive network tracing based loading aggregation method: 1) Start from a back-feeding source (usually a transformer), trace down all the network components it supplies, until the end of the tree structure is reached; 2) When returning to the source, the tracing method sums up the loading current at each network component and if applicable, compared with its corresponding limit; 3) The available capacity of a source can be calculated after the tracing goes back to the source. The available capacity factor of a source can be defined as: ( S LD ) S sum acf = (2) S Where S is the source capacity, LD sum is the present loading level of the source. E. Path Selection at T-Node A T-node is defined as the connection point of a lateral in a feeder. If the isolated network has T-nodes, its pre-restoration tree structure will define the isolation switch as the root and the potential back-feeding tie switches as the termination end. Suppose that both of the two downstream branches of a T- node may be back-fed, the algorithm has to choose one out of the two, otherwise a circuit loop will be generated in the postrestoration network. If the two downstream branches are to be back-fed from the same source, the branch with higher loading capability (absolute value) all the way to the source is chosen; otherwise, either the source of the higher available capacity is chosen, or the source with the lower available capacity factor S acf (as computed according to Equation 2) is chosen. A back-feeding path is defined as the connected circuit component from the back-feed source to the to-be-closed tie switch. F. Single-Path and Multi-Path Restoration If a source can provide the restoration power over a single path to an out-of-service load zone, the restoration is called a single-path restoration. Otherwise, the out-of-service load zone may have to be split into two or more load zones to be back-feed, and the scenario is named as multi-path restoration. Both single-path and multi-path restorations may have to shed load in case the back-feed source capacity or feeder components loading capability is not sufficient. III. THE ALGORITHM As shown in Figure 2, the algorithm starts with a backfeeding isolation switch search. This search is done on the pre-fault network s tree structure, with the tripped breaker as the root. The search traces down the tree, finds the most downstream switch that passed the fault current, and names it the forward-feed isolation switch. It then traces down further Figure 2: Flowchart of the Algorithm

4 4 for the first layer of downstream switches, and names them the back-feed isolation switches SW iso [i]. Then, the algorithm applies the following recursive steps: 1. For each isolation switch SW iso [i], the algorithm traces downstream and finds the first T-node (multi-connection load node) L mc, which is the load node that connects with more than one downstream switch; 2. After finding the first L mc node, the following judgment is made: if one or more single back-feed path exists to restore all loads in the particular isolated network by itself, then the back-feed path with the maximum available capacity will be selected; the restoration algorithm then returns to the Step 1 to process the next isolation switch. Otherwise, when no single back-feed path has sufficient capacity, the algorithm searches all the immediate downstream switches that connect to the L mc and stores them in vector SW other []; 3. The algorithm continues to trace all the downstream closed switches between the first L mc and a downstream tie-switch or a downstream second L mc node and stores them in vector SWs[]; 4. By opening each switch in SWs[], the algorithm determines whether the network will be divided into two sub-networks, in which out-of-service loads can be restored respectively. If so, the switch that can best balance the loading levels of the back-feeding sources will be selected as the switch that should be opened in the final restoration strategy; 5. Otherwise, in the case that no such switch in SWs[] can be found, for each switch in SW other [], i.e., SW other [j], if the number of its downstream tie-switches is greater than 1, which means that another L mc node exists in the downstream of this SW other [j], switch SW other [j] is treated as an isolation switch in SW iso, and the algorithm goes back to Step 1; 6. If no switch in SWs[] can be found to have more than one downstream tie-switches, the downstream tie-switch will be searched instead, and the out-of-service loads will be restored from the alternative back-feed path that connects to the tie-switch up to switch SW iso [i] or SW other [j] by closing the tie-switch. If the algorithm stops before it can reach switch SW iso [i] or SW other [j] because validation check didn t pass, the algorithm will goes back to Step 1 or 2; 7. If more than one path can restore up to its corresponding SW other [j], the one with more remaining capacity at its SW other [j] will be selected, and the restoration will proceed upstream via L mc to the corresponding SW iso [i]; 8. If the restoration can reach SW iso [i] and still have remaining capacity, it will start from L mc and try to restore any loads downstream of L mc that is still out-ofservice (Step 6 stops before reaching SW other [j]). IV. SOLUTION EXAMPLES These examples are for algorithm validation and capability demonstration only. They do not represent any physical circuit (for example, the feeder breakers are used as tie switches, and loading capability of feeder conductors and switching devices are assumed sufficient enough so loading violation checks on these devices are not necessary). However, when a detailed network model is available, the algorithm can be easily modified to include loading violation checks of any branch devices. Figure 3 shows a single-path full restoration example, where a fault at T-node L3 must be isolated by opening a forward-feed isolation switch R3 and two back-feed isolation switches R6 and R10. In this example, back-feed sources S3 and S4 both have sufficient capacity to pick up the out-ofservice load on their corresponding restoration path and each tie switch R9 and R12 can be closed to achieve the restoration. The post-restoration circuit topology is shown in Figure 3(b). (a) Normal Topology (b) Post-Restoration Topology Figure 3: Single-Path Restoration Example Figure 4 shows a multi-path full restoration example, where a fault at load node L1 must be isolated by a forward-

5 5 feed isolation switch (in this case no forward restoration is required) R1 and a back-feed isolation switch R2. In this example, none of the back-feed sources S2-S5 can completely pick up all the loads that are left unserved after fault isolation. Hence the algorithm splits the network into two parts (as in Step 4 above by opening R13 and the out-of-service load is restored by closing both R9 and R12 (from both S3 and S4). The post-restoration circuit topology is shown in Figure 4(b). (a) Normal Topology (a) Normal Topology (b) Post-Restoration Topology Figure 5: Multi-Path Partial Restoration Example V. CONCLUSIONS (b) Post-Restoration Topology Figure 4: Multi-Path Full Restoration Example Figure 5 shows an extreme example where the splitting of the out-of-service load zones is still not enough. Following the fault at load L1, and its isolation by opening R1 and R2, none of the backfeed sources can pick up the out-of-service loads completely or even partially without violating the current capacity limits of those sources. Load L5 has to be shed in order to restore power to as many out-of-service loads as possible. The post-restoration circuit topology is shown in Figure 5(b). Note that the out-of-service load zone has to be split into three portions, according to the algorithm. This paper presents a deterministic algorithm that identifies a restoration strategy to restore the out-of-service load due to fault isolation while ensuring that the postrestoration network has a valid configuration. The algorithm is based on the concepts of network tracing and it supports both single-path and multi-path restoration. In case the network components are too stressed and even the multi-path restoration cannot restore all the out-of-service loads, the algorithm tries to shed minimal load while restoring as many other loads as possible. Application examples show that the algorithm can produce appropriate back-feed switching strategies for any network topology. VI. ACKNOWLEDGMENT The author would like to thank ABB Corporate Research for the funding support of this research. VII. REFERENCES [1] [2] G. Ockwell, Implementation of Network Reconfiguration for Taiwan Power Company, IEEE PES General Meeting, [3] D.M. Staszesky, D. Craig, C.Befus, Advanced Feeder Automation Is Here, IEEE Power & Energy Magazine, Sept./Oct [4] J. Fan, X. Zhang, Feeder Automation within the Scope of Substation Automation, Power System Conference and Exposition, Nov [5] V. S. Devi, and G. Anandalingam, Optimal Restoration of Power Supply in Large Distribution Systems in Developing Countries, IEEE Transactions on Power Delivery, Vol. 10, NO. 1, January 1995.

6 6 [6] J. S. Wu, K. L. Tomsovic, and C. S. Chen, A Heuristic Search Approach to Feeder Switching Operations for Overload, Faults, Unbalanced Flow and Maintenance, IEEE Transactions on Power Delivery, Vol. 6, No. 4, Oct [7] T. Taylor and D. Lubkeman, Implementation of Heuristic Search Strategies for Distribution Feeder Reconfiguraiton, IEEE Transactions on Power Delivery, January [8] Y. Moon, B. Cho, H. Park, H. Ryu, B. Ha, and S. Lim, Fault Restoration Algorithm Using Fast Tracing Technique based on the Tree-Structured Database for the Distribution Automation System, IEEE PES Summer meeting, [9] C. Liu, S. J. Lee, and S. S. Venkata, An Expert System Operation Aid for Restoration and Loss Reduction of Distribution Systems, IEEE Transactions on Power Systems, Vol.3 No. 2, May [10] C. Huang, Multi-objective Service Restoration of Distribution Systems Using Fuzzy Cause-Effect Networks, IEEE Transactions on Power Systems, Vol. 18, No. 2, May [11] Y. T. Hsiao and C. Y. Chien, Enhancement of Restoration service in Distribution Systems Using a Combination Fuzzy-GA Method, IEEE Transactions on Power Systems, Vol. 15, Nov [12] Y. Ke, Distribution Feeder Reconfiguration for Load Balancing and Service Restoration by Using G-Nets Inference Mechanism, IEEE Transactions on Power Delivery, Vol. 19, No. 3, July [13] J. Wu, A Petri-Net Algorithm for Multiple Contingencies of Distribution System Operation, IEEE Transactions on Power Systems, Vol. 13, No. 3, August [14] Y. Ke, Distribution Feeder Reconfiguration for Load Balancing and Service Restoration by Using G-Nets Inference Mechanism, IEEE Transactions on Power Delivery, Vol. 19, No. 3, July [15] Q. Zhou, D. Shirmohammadi, W. E. Liu, Distribution Feeder Reconfiguration for Service Restoration and Load Balancing, IEEE Transactions on Power Systems, Vol. 12, No. 2, May [16] K. N. Miu, H. D. Chiang, B. Yuan, G. Darling, Fast Service Restoration for Large-Scale Distribution Systems with Priority Customers and Constraints, IEEE Transactions on Power Systems, Vol. 13, No. 3, August [17] C. Chen, C. Lin, and H. Tsai, A Rule-Based Expert System with Colored Petri Net Models for Distribution System Service Restoration, IEEE Transactions on Power Systems, Vol 17, No. 4, Nov [18] Paul W. Cox, Self-Healing Networks: Performance Improvement by Automated Switching Algorithm, CIRED Seminar 2008: SmartGrids for Distribution, paper No.0006, Frankfurt, June 2008 optimization, power system reliability analysis, the application of artificial intelligent techniques in power system control. James Stoupis (M 1991) is a Principal Consulting R&D Engineer in the Power Technologies Department for ABB s US Corporate Research Center located in Raleigh, North Carolina. Jim has been employed at USCRC for 12 years, and his research has been focused in the areas of distribution and feeder automation, wireless communications, power system protection and control, and event detection and classification. VIII. BIOGRAPHIES Zhenyuan Wang (M 2000) joined ABB US Corporate Research Center in Raleigh, North Carolina in 2000, where he is currently a Principal Consulting R&D Engineer. His research interests include electric power equipment condition monitoring/assessment/diagnosis, system monitoring, control and automation for a smart grid. His experiences include asset management IT applications in the electric power industry, power system transient analysis, substation/distribution automation, and data integration/warehousing/mining applications. Vaibhav Donde (M 2001) joined ABB US Corporate Research Center in Raleigh, North Carolina in 2006, where he is currently a Consulting R&D Engineer. Prior to joining ABB, he had a postdoctoral appointment at Lawrence Berkeley National Laboratory ( ). He holds Ph.D. (2004) and M.S. (2000) degrees, both in electrical engineering from the University of Illinois at Urbana-Champaign and a B.E. degree (1998) in electrical engineering from V.J.T.I., Mumbai, India. He has worked with TATA Consulting Engineers, Mumbai ( ). His technical interests include power system analysis, modeling and simulation, power distribution systems and automation, hybrid dynamical systems and nonlinear control. Fang Yang (M 2007) joined ABB US Corporate Research Center in Raleigh, North Carolina in Her research interests include distribution automation and voltage/var