Projeto de redes de distribuição de energia com incertezas na evolução da carga utilizando algoritmos meméticos

Abstract

Electric power distribution systems have highlighted economic and social importance since they are responsible for suplying 85% of the country population. This unquestionable importance, together with the high budget usually required for installing those systems, justify the use of optimisation techniques in the distribution system design. Optimised distribution networks provide better allocation of the financial resources, reducing total installation costs and power losses. The distribution network design is intrinsically complex due to the discrete characteristic of the search space and the nonlinearities involved. Therefore the traditional optimisation algorithms such as deterministic continuous methods become unviable. An optimisation algorithm class that has been recently used in the optimisation of such systems is the evolutionary algorithms due to their flexibility, robustness and capacity of adaptation. Memetic Algorithms, which are a hybridization from an evolutionary algorithm and a local search method, emerge as an interesting option since they are a powerful tool for obtaining and refining global and local optima. Memetic algorithms proposed here are the hybridization from a Clonal Selection Algorithm, which is able to find a set of optimal and sub-optima solutions, and a local search method, which is based on the the T-norm network metric, that generates random networks at pre-defined random distances. The set of optimal solutions mapped by the memetic algorithms are very useful for dealing with an important aspect of the eletric distribution network design - the load evolution uncertainties. They are treated by means of a multiobjective sensitity analysis. Monte Carlo Simulation is used to provide random possible load scenarios. The set of non-dominated solutions found by the sensitivity analysis helps the designer in his choice of the best network. Possible changes in load conditions are taken into account and robust networks, which can handle with design uncertainties, are obtained. An important result gained here is the possibility of using a highly robust networks spending just a few money more than that which would be used for the construction of a network designed for the medium load scenario.