基于动态分配邻域策略的分解多目标进化算法

Decomposition multi-objective evolutionary algorithm based on dynamic allocation neighborhood strategy

邻域规模是影响分解多目标进化算法性能的重要因素之一,传统分解多目标进化算法通常对计算复杂度不同的子问题分配相同的邻域规模,算法运行效率受到限制。针对以上问题,提出了一种基于动态分配邻域策略的分解多目标进化算法(MOEA/D-SD):首先,在收敛性方向和多样性角度上评估个体的进化状态;其次,根据进化状态动态调节邻域大小,为每个子问题分配合适的邻域规模,从而提高算法的收敛性和解集整体质量。将所提算法与MOEA/D和MOEA/D-GR算法在ZDT和DTLZ系列测试函数上进行性能对比。仿真实验结果表明:MOEA/D-SD算法的收敛性能明显提升,算法资源分配更加合理,所求解集整体质量有所提升。

The neighborhood size is an important factor to affect the performance of the decomposition multi-objective evolutionary algorithm(MOEA/D).MOEA/D usually allocates the same neighborhood size to subproblems with different computational complexity,so the efficiency of the algorithm is limited.Based on this problem,this paper proposes a decomposition multi-objective evolutionary algorithm based on dynamic allocation neighborhood strategy(MOEA/D-SD).Firstly,the evolutionary status of individuals from the perspective of convergence direction and diversity are evaluated.Then,the size of the neighborhood size according to the evolution state is adjusted dynamically and an appropriate neighborhood size to each sub-problem is assigned in order to improve the convergence of the algorithm and the overall quality of the solution set.The test results on ZDT and DTLZ test suite show that,comparing to MOEA/D and MOEA/D-GR algorithms,the convergence performance of MOEA/D-SD algorithm is significantly improved,the algorithm resource allocation is more reasonable,and the overall quality of the solution set is improved.