多目标进化算法中基于聚集距离调整的分布性保持方法

CROWDING DISTANCE ADJUSTMENT-BASED DISTRIBUTION PROPERTY MAINTENANCE STRATEGY IN MOEAS

在经典的非支配排序遗传算法中,基于聚集距离的种群维护策略并不能很好地保持解集的分布性。提出一种改进的基于聚集距离调整的分布性维护策略,根据邻近个体的聚集距离大小关系,保留分布较好的个体。与经典算法NSGA-Ⅱ,PESA-Ⅱ和小生境方法进行比较,实验结果表明,提出的分布性维护策略能较大程度提高分布性,并保持较好的收敛性。

In classical non-dominated sorting genetic algorithm, population maintenance strategies based on crowding distance can not well maintain the distribution property of its solutions. We propose an improved distribution property maintenance strategy. It is based on crowding distance adjustment and reserves the well-distributed individual solutions according to the size relation of the crowing distance of adjacent indi- viduals. Compared with classical NSGA- Ⅱ, PESA- Ⅱ and NICHE, the experimental results demonstrate that the proposed distribution proper- ty maintenance strategy can improve the distribution property to a greater extent and keep better convergence at the same time.