Leader recommend operators selection strategy for a multiobjective evolutionary algorithm based on decomposition

Abstract

The decomposition-based multi-objective evolutionary algorithm (MOEA/D) shows strong performance in solving complex multi-objective problems (MOPs), wherein evolutionary operators have great influence on the search ability. Appropriate operators can greatly improve the performance of MOEA/D. However, one omnipotent operator usually cannot handle all different MOPs very well. According to the characteristics of the problems, the operators that adapt to the problems are also different. This paper proposes an adaptive operator selection strategy called Leader Recommend Operator Selection (LROS). We construct an operator pool consisting of Simulated Binary Crossover (SBX) and three operators of Differential Evolution (DE). Meanwhile, we divide the entire evolutionary process into several stages, and each stage is divided into Part 1 and Part 2. In Part 1 of each stage, we use a subset of individuals to test the performance of each operator in the operator pool and choose the best performance. In Part 2, we adopt the best performing operator selected in Part 1 to generate offspring, meanwhile we verify the actual quality of these offspring to decide in the next stage whether to continue using this operator to generate offspring or to choose a more suitable operator. Experimental results demonstrate the effectiveness of LROS.