Publicação

Many-objective optimization using differential evolution with variable-wise mutation restriction

Detalhes bibliográficos
Resumo:	In this paper, we propose an evolutionary algorithm for handling many-objective optimization problems called MyO-DEMR (many-objective differential evolution with mutation restriction). The algorithm uses the concept of Pareto dominance coupled with the inverted generational distance metric to select the population of the next generation from the combined multi-set of parents and offspring. Furthermore, we suggest a strategy for the restriction of the difference vector in DE operator in order to improve the convergence property in multi-modal fitness landscape.We compare MyO-DEMR with other state-of-the-art multiobjective evolutionary algorithms on a number of multiobjective optimization problems having up to 20 dimensions. The results reveal that the proposed selection scheme is able to effectively guide the search in high-dimensional objective space. Moreover, MyO-DEMR demonstrates significantly superior performance on multi-modal problems comparing with other DE-based approaches.
Autores principais:	Denysiuk, Roman
Outros Autores:	Costa, Lino; Espírito Santo, I. A. C. P.
Assunto:	Multiobjective optimization Multiobjective evolutionary algorithms Performance assessment
Ano:	2013
País:	Portugal
Tipo de documento:	comunicação em conferência
Tipo de acesso:	acesso restrito
Instituição associada:	Universidade do Minho
Idioma:	inglês
Origem:	RepositóriUM - Universidade do Minho

Descrição
Resumo:	In this paper, we propose an evolutionary algorithm for handling many-objective optimization problems called MyO-DEMR (many-objective differential evolution with mutation restriction). The algorithm uses the concept of Pareto dominance coupled with the inverted generational distance metric to select the population of the next generation from the combined multi-set of parents and offspring. Furthermore, we suggest a strategy for the restriction of the difference vector in DE operator in order to improve the convergence property in multi-modal fitness landscape.We compare MyO-DEMR with other state-of-the-art multiobjective evolutionary algorithms on a number of multiobjective optimization problems having up to 20 dimensions. The results reveal that the proposed selection scheme is able to effectively guide the search in high-dimensional objective space. Moreover, MyO-DEMR demonstrates significantly superior performance on multi-modal problems comparing with other DE-based approaches.