Feeder Load Balancing Using Genetic Algorithms and Artificial Neural Network

Abstract

Low voltage power distribution system problems such as system planning, energy loss minimization and restoration usually involve proper load balancing or network reconfiguration procedures. To achieve an appreciable level of load phase balance, feeder reconfiguration using appropriate switching control strategy such as: Simulated Annealing, Tabu Search, Particle Swarm Optimization, and heuristic algorithms are viable preferences. However, the systematic solution to load phase balancing can be greatly enhanced optimally through implementation of an appropriate combinatorial optimization procedure such as Genetic Algorithms and Artificial Neural Network. Accordingly, this paper presents a genetic algorithms procedure to enhance the load phase balancing optimization and then train an artificial neural network to automate the reconfiguration of the distribution network loads, thus ensuring an optimal phase balancing in the system. An Intel® 2.0 GHz, 4GB RAM HP255 computer-based MATLAB® 14 was used for the neural network training, testing, and the implementation of the genetic algorithms. The outputs of the algorithms are the switching sequence for a balanced network. The parameters ΔIph (max - min) and Δ(Iph – Imax) which is the maximum difference between the phase currents, which are ideally zero if there are no imbalances in the network, shows considerable improvement in the balancing when compared with other literatures. This work presents the application examples of the proposed methods using real test data.