基于目标空间分解的自适应多目标进化算法

An adaptive multi-objective evolutionary algorithm directed by objective space decomposition

针对基于分解的多目标进化算法(MOEA/D)个体与子问题的匹配问题,在分析MOEA/D的进化规律的基础上,提出了一种基于目标空间分解的自适应多目标进化算法(MOEA/OSD)。该算法采用以测试问题的参考点为起点的均匀权重向量分解目标空间,根据个体信息动态选择适合的子问题,并使用辅助向量的方法弥补分解方法的不足。对比实验结果表明,MOEA/OSD拥有较好的收敛性和分布性,采用不同的分解方法均能搜索到最优解,且具有较好的收敛速度。

Aiming at solving the matching problem of individual and sub-problem of the multi-objective evolutionary algo- rithm based on decomposition （MOEA/D）, the paper proposes an adaptive multi-objective evolutionary algorithm directed by objective space decomposition （MOEA/OSD） based on the evolution analysis of the MOEA/D. The MOEA/OSD decomposes an objective space by even spread weight vectors whose start points are the reference points, chooses a suitable sub-problem by using the information of individuals, and uses auxiliary weight vectors to compensate for the limitations of the decomposition approaches. The experimental results demonstrates that the MOEA/OSD could not only balance the convergence and diversity effectively but also approach the optimal solution by applying different decomposition approaches, and has a better convergence speed.