基于存档和权值扩展的大规模多目标优化算法

Large-Scale Multi-Objective Optimization Algorithm Based on Archive and Weight Extension

由于不同目标之间相互冲突且搜索空间巨大,现有大规模多目标优化算法的综合性能尚存在较大改进空间.为合理均衡算法的搜索效率与搜索质量,提升算法的综合性能,本文提出一种基于存档和权值扩展的大规模多目标优化算法(LSMOEA-AWE).该算法总体采用进化计算框架,基于大规模决策变量与小规模权值变量之间的问题转换进行求解.其核心是在进化过程中选取高质量代表性解及其对立点构建存档高效引导种群的进化方向,并引入权值扩展策略逐步扩大算法的搜索空间,在确保算法搜索效率的同时,提升搜索质量.为验证LSMOEA-AWE的有效性,将其与6个先进的大规模多目标优化算法在最新的大规模多目标基准测试问题集LSMOP上进行对比,实验结果表明LSMOEA-AWE对于大规模多目标优化问题的求解具有明显的竞争优势.

Due to the conflict between different objectives and the huge search space,the comprehensive performance of existing large-scale multi-objective optimization algorithms still has large room for improvement.In order to balance the search efficiency and search quality reasonably,and improve the comprehensive performance of the algorithm,this paper proposes a large-scale multi-objective optimization algorithm based on archive and weight extension(LSMOEA-AWE).LSMOEA-AWE adopts the framework of evolutionary computation,and the problem transformation scheme between the large-scale decision variables and the small-scale weight variables.LSMOEA-AWE consists of two essential components.One is the archive strategy based on the high-quality representative solutions and their opposite points in each iteration,which can effectively guide the rational evolutionary direction of the population.The other is the weight extension strategy,which can gradually expand the search space of the algorithm.In order to verify its effectiveness,LSMOEA-AWE is compared with six state-of-the-art algorithms in large-scale multi-objective optimization on the latest test suite LSMOP.The experimental results show that LSMOEA-AWE has obvious competitive advantages in solving large-scale multi-objective optimization problems.