基于方向向量采样的大规模多目标进化优化

Large-scale multi-objective evolutionary optimization based on direction vector sampling

针对在高维决策空间的大规模多目标优化中难以搜索产生有希望的后代个体,从而导致常规算法性能恶化的问题,提出了一种基于方向向量采样的大规模多目标进化算法.该算法首先选取一些接近理想点的优秀候选个体,然后根据候选个体分阶段创建方向向量,最后执行定向采样策略(收敛性相关采样策略和多样性相关采样策略)来生成子代个体,由此生成的子代个体性能优秀,且算法收敛加速.实验结果表明,与5种具有代表性的大规模多目标进化算法相比,该算法在多达2000个决策变量的大规模多目标优化测试中,仍具有较强的寻优能力.

To tackle the difficulty to search for promising offspring individuals in large-scale multi-objective optimization of high-dimensional decision space,which leads to the deterioration of the performance of conventional algorithms,a large-scale multi-objective evolutionary algorithm based on direction vector sampling is proposed.The algorithm first selects some good candidate individuals close to the ideal point,then creates direction vectors in stages,and executes directional sampling strategies,that is,convergence-related sampling strategy and diversity-related sampling strategy to generate offspring individuals.In this way,the offspring individuals with better performance are generated to accelerate the convergence of the algorithm.Experimental results show that compared with five representative large-scale multi-objective evolutionary algorithms,the proposed algorithm has retained strong optimizing ability on large-scale multi-objective optimization test problems with as many as 2000 decision variables.