Estudo comparativo entre metaheutísticas populacionais com tamanho da população variável

AUTOR(ES)
DATA DE PUBLICAÇÃO

2008

RESUMO

This work introduces four novel population-based heuristic algorithms, whose population size varies along the execution, which are aimed at solving problems of search and numerical optimization. These algorithms are extensions of the standard models of two metaheuristics recently proposed in the literature, which have been successfully applied in different fields. They are: Particle Swarm Optimization (PSO) and Differential Evolution (DE). In addition, these new algorithms are daptations of two other models proposed in the context of Genetic algorithms (GAs), namely, the Adaptive Population size GA (APGA) and Population Resize on Fitness Improvement GA (PRoFIGA). In order to empirically validate the proposed algorithms, their implementations are evaluated in terms of efficiency and effectiveness in three different case studies: optimization of benchmark numerical functions; prototype selection for data clustering; and training of feedforward neural networks. The results obtained in the benchmark functions optimization indicate gains, in terms of the ffectiveness issue, for time-varying population size models. Conversely, the results achieved by the time-varying population size models when dealing with the data clustering task have not shown gains in erformance. Finally, in the training of artificial neural networks, the novel algorithms could utperform the standard models in terms of effectiveness criterion, although the gains incurred were less expressive than those obtained in the first case study. Keywords: Optimization, Population-based Metaheuristics, Parameter Control, Evolutionary Computing, Particle Swarm Optimization, Differential Evolution, Data Clustering, Artificial Neural Netwo

ASSUNTO(S)

redes neurais - dissertaÇÕes heurÍstica (informÁtica) - dissertaÇÕes sistemas de informacao otimizaÇÃo matemÁtica - dissertaÇÕes

ACESSO AO ARTIGO

Documentos Relacionados