Multiobjective Differential Evolution Algorithm with Opposition-Based Parameter Control

Multiobjective evolutionary algorithms (MOEAs) often have several control parameters, and their performance is highly related to the parameters. A proper set of parameter values is useful for MOEAs in a particular application. This paper addresses the parameter control problem. Inspired by the observations in differential evolution (DE), we proposed a parameter control system using opposition-based learning (OBL). The proposed method contains three conditions which characterize the state of parameters at different evolutionary stages. It keeps good parameters for the current search stage. In case the parameters are bad, it uses OBL to accelerate the finding of good ones. The method is applied to a newly proposed multiobjective DE algorithm (MODEA) which does not control parameters. The resulting algorithm is tested on CEC 2009 test suite comparing with two other recently proposed MOEAs, namely GDE3 and MOEA/D. Experimental results show that the proposed method can significantly improve the performance of MODEA. Moreover, the resulting algorithm significantly outperforms GDE3 and MOEA/D.