基于极大极小关联密度的多目标微分进化算法

Multi-objective differential evolution algorithm based on max-min correlation density

采用微分进化方法求解多目标优化问题,为了改善解集分布性和提高算法收敛性,提出1种基于极大极小关联密度的多目标微分进化算法。该算法定义了极大极小关联密度。在严格遵守Pareto支配规则的基础上,给出了基于极大极小关联密度的外部档案集维护方法,从而避免或减少最终解集的多样性损失。1种自适应选择策略通过评价个体的关联密度来指导个体优劣的选择过程,在确保最优个体进入下一代种群的同时,尽可能使个体的选择覆盖更广泛的搜索空间。实验结果显示,与多目标均匀多样性差分进化(MUDE)、基于反对称的自适应混合差分进化(OSADE)和非劣排序遗传算法II(NSGA-II)等经典算法相比,该文算法在世代距离(GD)和空间(SP)性能指标上有更好的表现,具有更优的Pareto前沿分布性与收敛性。

In order to improve the distribution of solution set and the convergence of the algorithm for multi-objective optimization problems solving by differential evolution,a multi-objective differential evolution algorithm based on max-min correlation density is proposed.A max-min correlation density is defined.Observing the Pareto domination rule strictly,a maintenance method of external archives based on the max-min correlation density is given to avoid or reduce the loss of diversity of the final solution set.An adaptive selection strategy is designed to guide the selection process of the individual and ensure that the optimal individual enters the next generation of population by evaluating the correlation density of the individual,and the individual selection is covered in a wider search space as much as possible.The experimental results show that the proposed algorithm has better performance in generational distance(GD)and spacing(SP)performance criteria than other comparison algorithms,such as multi-objective uniform-diversity differential evolution(MUDE),opposition-based self-adaptive differential evolution(OSADE)and non-dominated sorting genetic algorithm II(NSGA-II),and has a better distribution and convergence of the Pareto frontier.