关键词: 大规模多目标优化, 协同进化, 决策变量分析, 主成分分析, 邻域自适应更新

Abstract:

Since there are large scale decision variables in large-scale multi-objective optimization problems (LSMOPs), the algorithm is difficult to converge and the distribution of the solution set is uneven. It is an effective way to classify the decision variables and optimize them respectively by analyzing the characteristics of the variables. However, there are some shortcomings, such as inaccurate variable classification and insufficient pertinence of variable processing. Therefore, a large-scale multi-objective optimization based on differential evolution with neighborhood adaptive strategy (NAS-MOEA) is proposed to solve LSMOPs. Firstly, by analyzing the dominant relationship of the disturbance solution, the mixed variables are divided into diversity variables and convergence variables to make the variable classification more accurate. Secondly, the principal component analysis of convergence variables is used to reduce noise and computational cost. The alternative evolution strategy of population and the neighborhood adaptive update operation of differential evolution are designed to improve the convergence in the process of population evolution. Experimental results show that the proposed algorithm has good performance in convergence speed and uniformity distribution of the solution set.

Key words: large-scale multi-objective optimization, cooperative coevolution, decision variable analysis, principal component analysis, neighborhood adaptive update