Research of Strategies of Maintaining Population Diversity for MOEA/D Algorithm

Abstract

In recent years, MOEA/D algorithm has been recognized by the industry for its inherent advantages in dealing with super multi objective optimization problems, and its application is also very extensive. However, MOEA/D algorithm also has the problem of lack of population diversity during the later stage of evolution, resulting in slow convergence speed. In this paper, it makes a research on the strategy of maintaining population diversity based on MOEA/D algorithm and proposes three population diversity maintenance strategies, namely SBX-DE operator competition, mutation probability adaptive modulation, and double-faced mirrors theory boundary processing. The experiments’ result shows that all of these three strategies can effectively improve the diversity of the MOEA/D algorithm in the late evolutionary population, and contribute to the convergence speed of the MOEA/D algorithm.