关键词: 差分进化, 多目标优化, 差分变异策略, 精英选择策略

Abstract: To improve the convergence and distribution of Multi-Objective Evolutionary Algorithms(MOEAs) in dealing with large-dimensional Multi-objective Optimization Problems (MOPs), a Self-adaptive Differential Evolution Multi-objective Optimization (SDEMO) was proposed. Based on the model of Nondominated Sorting Genetic Algorithm II (NSGA-II), the elitist selection strategy and the crowding distance calculation in the model of SDEMO were improved to achieve better convergence performance. In addition, new mutation strategy as well as new parameter control strategy of Differential Evolution (DE) algorithm were also presented according to the characteristics of MOPs. SDEMO was compared to 6 state-of-the-art MOEAs on benchmark test problems. Simulation results showed that SDEMO could ensure good convergence while had uniform distribution and wild coverage area for obtained Pareto optimum solution. It had obvious advantages than other algorithms, especially, applied to solving large-dimensional MOPs.

Key words: differential evolution, multi-objective optimization, differential mutation strategy, elitist selection strategy, crowding density estimation