A Regularized Evolutionary Algorithm for Constrained Optimization
Carlos Cristiano Hasenclever Borges1; Raul Fonseca Neto1; Sam Ould Mohamed El Hacen1
1 Universidade Federal de Juiz de Fora
doi:10.20906/CPS/CILAMCE2015-0514
Resumo
Penalty techniques in conjunction with evolutionary algorithms applied for constrained optimization problems are usually adopted. However, this kind of modeling suffers with the difficulty to set adequate values for the penalization parameters. Dynamic and adaptive models are an attempt aiming to circumventing this drawback commonly used. Despite the quality obtained in constrained optimization using dynamic or adaptive models both suffers with the greediness phenomenon. When the problem has this characteristic the solution tends to be attracted to the infeasible region due to the balance between the objective function and penalty terms. In this work a regularized formulation is proposed in the context of a optimization evolutionary algorithm framework with the perspective to diminish the greediness effect. Tests in mechanical and structural optimization problems are carried out to evaluate the proposed strategy and its performance in relation to a similar non-regularized method.
Palavras-chave: Constrained Optimization; Evolutionary Computation; Regularization