Service Restoration in Large-Scale Distribution Systems Considering Three Levels of Priority Customers Leandro T. Marques, Alexandre C. B. Delbem and Jo˜ ao Bosco Augusto London Junior University of S˜ ao Paulo, S˜ ao Carlos, Brazil leandrotolomeu@gmail.com, acbd@icmc.usp.br, jbalj@sc.usp.br Marcos Henrique M. Camillo COPEL Distribuic ¸˜ ao S/A Londrina, Brazil marcos.camillo@copel.com Abstract—Service restoration in contingency situations is one of the most critical problems in the context of distribution system operation. It is a problem with multiple objectives and multiple constraints whose solution must be gotten as soon as possible. Besides, due to the existence of customers with higher priority of supply (e.g., big industries and hospitals), it is necessary to prioritize service restoration to these loads. However, it is important to highlight that these priority customers have different priority levels, that also must be considered during the determination of a feasible service restoration plan. In this paper, a methodology based on multi-objective evolutionary algorithm is proposed for solving service restoration problem in large-scale distribution system taking into account the existence of three levels of priority customers. Simulations results have shown the proposed methodology is able to find suitable service restoration plans for large-scale real distribution systems (from 3,860 to 30,880 buses) with relatively soft computing without requiring any network simplification. Index Terms—Large-Scale Distribution Systems; Service Restoration; Priority Customers; Multi-Objective Evolutionary Algorithm. I. I NTRODUCTION Distribution Systems (DSs) must operate continuously, eco- nomically and reliably. However, as they are generally oper- ated in radial configurations (although they are structured in meshes), steady-state faults in DSs can produce large out-of- service areas. As a consequence, to reduce service interruption frequency and duration, the load buses and lines of DSs are grouped in blocks (called sectors), which are separated by Normally Open (NO) and Normally Closed (NC) switches. Thus, when a steady-state fault occurs, it is possible to isolate the faulted sectors of the network and perform load exchange between substations and feeders by network reconfiguration. Network reconfiguration is the process of altering the topo- logical structure of DSs by opening NC switches and closing NO switches. When network reconfiguration is applied to the service restoration (SR) problem (which emerges after the faulted areas have been identified and isolated), usually the objectives are to minimize both number of out of service sectors and number of switching operations without violating the radiality and operational (limits for the node voltage, network loading, and substation loading) constraints. In the course of SR, it is important to consider that in any DS there are always some loads which are of the highest priority (e.g., hospitals, steel industries, subway stations, big industries, big supermarkets etc.) [1], [7]. Hereafter these cus- tomers will be called Priority Customers (PCs). However, note that a hospital has a need of supply higher than an industry. Consequently, SR plans must be designed to prioritize supply to PCs considering their levels of priority. Therefore, besides to minimize the number of out of service sectors and the number of switching operations, the prioriti- zation of supply to PCs, according to their priority levels, is also an objective of the SR problem. Thus, the SR problem can be classified as a multi-objective and multi-constraint optimization problem. Furthermore, the SR problem is a combinatorial problem, due to the large number of switching elements, and must be solved as fast as possible to ensure customer’s satisfaction and avoid penalizations for the utilities. Because of the characteristics mentioned above, the SR problem is computationally complex (NP-complete) [7], which in practice means that there is no known mathematical com- puter algorithm that can guarantee the optimal solution in a reasonable runtime (and perhaps no one will be able to design such algorithm). That’s why meta-heuristic-based methods have been widely explored and proposed. To treat SR problem, many methods have been suggested. Table I summarizes the main features of the most recent and relevant of them (Table I listed the methods in a chronological order). Observe that the most of the methods in the literature are validated in small networks, which do not correspond to real DSs (with hundreds of buses and switches) [2]– [9]. Among the methods listed in Table I, it is important to highlight the methods proposed in [1] and [10], which prioritize SR to PCs considering multiple levels of PCs. These methods are based on Multi-Agent Systems (MASs) and on Expert Systems (ESs), respectively. MASs-based methods have the inconvenience of being network infrastructure de- pendent and require, for example, communication channels, remotely/automatically controlled switches, data acquisition and monitoring, and others. On the other hand, ESs are