Abstract: To improve the convergence and distribution of Multi-Objective Evolutionary Algorithms (MOEAs) on solving high dimensional multi-objective optimization problems (MOPs),a cellular multi-objective genetic algorithm based on multi-strategy differential evolution was proposed.Through analyzing the different differential evolution strategies,a selection operator of multi-strategy differential evolution based on the cellular model was proposed.Aiming at the defects of the present crowding diversity measure,a new crowding diversity measure method based on entropy was introduced and the replacement policy was also improved.Through testing 12 benchmark functions,the proposed algorithm was proved to have better convergence and diversity than NSGA-Ⅱ,MOCell and CellDE,and the coverage rate for solution was also improved,which especially for solving high dimensional multi-objective optimization problems.

Key words: cellular model, multi-strategy differential evolution, multi-objective optimization, crowding diversity measure, replacement policy

摘要： 针对现有的多目标进化算法在求解复杂的多目标优化问题时收敛性不佳和解的分布性差等问题,提出一种基于多策略差分进化的元胞多目标遗传算法。通过分析不同差分进化模式的优劣,结合元胞模型,定义了一种多策略差分协同进化的选择算子;针对当前拥挤距离评估方法存在的缺陷,引入一种基于熵的拥挤距离评估方法,同时改进了替换策略。通过12个标准测试函数进行测试,证明了新算法相对于非支配排序遗传算法、元胞多目标遗传算法和混合元胞遗传算法,不仅具有更好的收敛性和多样性,而且在解的覆盖率上得到了一定程度的提高,尤其适合于高维复杂多目标优化问题的求解。

关键词: 元胞模型, 多策略差分进化, 多目标优化, 拥挤距离评估, 替换策略