A multi-operator-based large-scale multi-objective optimization evolutionary algorithm

Abstract

Large-scale multi-objective optimization brings significant challenges for offspring generation, due to its multiple conflicting objectives and huge searching space. For most of large-scale multi-objective optimization evolutionary algorithms (LSMOEAs), the existing offspring generation mechanism often fails to converge to true Pareto front rapidly. To remedy this issue, this paper proposes an algorithm, named MOLMOEA, in which multiple generation operators are involved. The first operator conducts orientations to help solutions jumping out of local optimum regions, and the second operator employs the main idea of competition to improve diversity of solutions. Furthermore, a comprehensive indicator is proposed to measure the quality of solutions. The performance of MOLMOEA is validated against five mainstream algorithms on a set of large-scale multi-objective optimization problems (LSMOPs) benchmark, and the performance of compared algorithms is measured by IGD. The experimental results demonstrate that MOLMOEA achieves superior performance on LSMOPs.