Power distribution network reconfiguration by evolutionary integer programming

This paper presents and analyses new metaheuristics for solving the multiobjective (power) distribution network reconfiguration problem (DNRP). The purpose of DNRP is to minimize active power loss for single objective optimization, minimize active power loss and minimize voltage deviation for multi-objective optimization. A non-redundant integer programming representation for the problem will be used to reduce the search space size as compared to a binary representation by several orders of magnitudes and represent exactly the feasible (cycle free, non-isolated node) networks. Two algorithmic schemes, a Hybrid Particle Swarm Optimization - Clonal Genetic Algorithm (HPCGA) and an Integer Programming Evolution Strategy (IES), will be developed for this representation and tested empirically. Conventional algorithms for solving multi-objective DNRP are converting the multiple objective functions into a single objective function by adding weights. However, this method cannot capture the trade-offs and might fail in case of a concave Pareto front. Therefore, we extend the HPCGA and IES in order to compute Pareto fronts using selection procedures from NSGA-II and SMS-EMOA. The performance of the methods is assessed on large scale DNRPs.