Configuração de sistemas de distribuição usando um algoritmo genético sequencial.

Abstract

The Distribution Network Configuration Problem is stated to be solved by using Genetic Algorithms (GAs). The formulation includes network reconfiguration, service restoration, and expansion planning. A fitness function comprising those three approaches, a procedure to automatically adjust its parameters and a genetic memory which mitigates the overall fitness calls are presented. However, despite this general formulation, the main focus of the thesis is in developing new techniques so as to configure large networks using GAs in a feasible time. Accordingly, two encodings are presented: Subtractive and Additive. They share the advantage of representing only radial and connected networks, the fact that adopt a logic whose interpretation is sequential, and do not demanding any specific genetic operator. Their main drawback is the complexity of decodification algorithms, so specific data structures and operators were proposed in order to simplify the decodification process: L, L’, XOR, XOR’, R, R’, IXOR and IXOR’. The results on five test systems available in the configuration literature show that these encodings are effective in improving the performance of GAs. Although Subtractive and Additive characterize a contribution to the state of the art, it is considered as the main innovation of the thesis the formulation of a Sequential GA. This algorithm shall use some sequential encoding and differs from the conventional GA combining the control of search space size with a heuristic optimization criterion. In order to validate the Sequential GA, it was used the largest three systems of the five used previously. Consistent improvements in performance were observed, taking as reference even the results obtained by Subtractive and Additive with the GA.