Evolução diferencial intervalar: uma abordagem baseada em decomposição estrutural de problemas de otimização global
Visualizar/ Abrir
Data
2019Autor
Cassenote, Mariane Regina Sponchiado, 1993-
Metadata
Mostrar registro completoResumo
Resumo: Algoritmos de Evolução Diferencial (DE) têm se mostrado promissores para abordagem de problemas de otimização numérica global com restrições. Muitas variantes recentes de DE são aplicadas a otimização caixa-preta, em que a estrutura analítica da instância é desconhecida. Neste trabalho é apresentado um algoritmo InDE (do inglês Interval Differential Evolution), que explora as informações de uma decomposição estrutural da instância durante o processo de otimização. As restrições e a função objetivo da instância são decompostas em operadores e variáveis originais e auxiliares. Então, a estrutura da instância é codificada em um hipergrafo cujas arestas representam os operadores e vértices representam as variáveis. Uma decomposição Epífita é aplicada sobre o hipergrafo, permitindo que um subconjunto de variáveis críticas seja extraído. Os operadores evolutivos intervalares do InDE são aplicados somente sobre esse subconjunto de variáveis, atribuindo a cada variável um intervalo de valores reais. Os intervalos das demais variáveis da instância são dados por propagação através do hipergrafo de restrições. O otimizador implementado utiliza diversas abordagens adaptativas estado-da-arte em otimização baseada em meta-heurísticas. Os experimentos indicam que a utilização de informação estrutural melhora significativamente o desempenho do algoritmo DE. Palavras-chave: Evolução Diferencial, Otimização Global, Decomposição Estrutural, Otimização Intervalar. Abstract: Differential Evolution (DE) algorithms have shown to be promising for tackling numerical constrained global optimization problems (NCOPs). Many recent variants of DE are applied to black-box optimization, in which the analytical structure of the instance is unknown. In this work it is presented an Interval Differential Evolution (InDE) algorithm that explores the information of a structural decomposition of the NCOP instance during the optimization process. The constraints and objective function are decomposed into operators, and original and auxiliary variables. Then, the structure of the instance is encoded in a hypergraph whose edges represent the operators and vertices represent the variables. An Epiphytic decomposition is applied to the hypergraph, allowing a subset of critical variables to be extracted. InDE's evolutionary interval operators are applied only on this subset of variables, assigning to each variable an interval of real values. The intervals of others variables of the instance are given by propagation through the constraint hypergraph. The implemented optmizer uses several state-of-the-art adaptive approaches of metaheuristic-based optimization. The experiments indicate that the use of structural information significantly improves the performance of the DE algorithm. Keywords: Differential Evolution, Global Optimization, Structural Decomposition, Interval Optimization.
Collections
- Dissertações [355]