基于粒子记忆体的多目标微粒群算法*

针对多目标微粒群算法(MOPSO)解的多样性分布问题，提出一种基于粒子记忆体的多目标微粒群算法(dp-MOPSO)。dp-MOPSO算法为每个微粒分配一个记忆体，保存寻优过程中搜索到的非支配pbest集，以避免搜索信息的丢失。采用外部存档保存种群搜索到的所有Pareto解，并引入动态邻域的策略从外部存档中选择全局最优解。利用几个典型的多目标测试函数对dp-MOPSO算法的性能进行测试，并与两种著名的多目标进化算法m-DNPSO、SPEA2进行比较。实验结果表明，dp-MOPSO算法可以更好地逼近真实Pareto沿，同时所得Pareto解分布更均匀。

New multi-objective particle swarm optimization based on extended individual memory

To deal with the problem of diversity distribution of solution in multi-objective particle swarm optimization,this paper proposed a diversity pbest based multi-objective particle swarm optimization algorithm (dp-MOPSO).In dp-MOPSO,allocated each particle an individual memory to save the solution set of non-dominated pbest which were fiound in the searching process, avoiding the loss of searching information.Used an external archive to save all the Pareto solutions, and introduced the dynamic neighborhood strategy to select the global optimal solution from the external archive.Tested several multi-objective benchmark functions for comparing the performance of dp-MOPSO with two famous multi-objective evolutionary algorithm m-DNPSO and SPEA2.The results show that dp-MOPSO converges to the true Pareto front more closely, and also all the Pareto solutions are well-distributed.