Experimental comparison of different differential evolution strategies in MOEA/D

MOEA/D is a well-known optimization algorithm in dealing with complex multi-objective problems. It employs a simple differential evolution strategy to generate offspring individuals. However, duo to the sensibility to the parameter setting in differential evolution strategy, MOEA/D performs poor in certain problems. To understand the influences of different DE strategies, this paper tries to investigate the overall performance of MOEA/D with different DE strategies. The experiment results demonstrate that DE/current-to-rand/1 strategy performs the best in all test problems.