基于粒子记忆体的多目标微粒群算法  

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

作　　者：章国安[1,2] 周超[1] 周晖[1] 

机构地区：[1]南通大学电子信息学院,江苏南通226019 [2]东南大学移动通信国家重点实验室,南京210096

出　　处：《计算机应用研究》2010年第5期1665-1668,共4页Application Research of Computers

基　　金：国家"863"计划资助项目(2007AA01Z330);江苏省高校自然科学重大基础研究资助项目(07KJA51007);江苏省教育厅"青蓝工程"资助项目;东南大学移动通信国家重点实验室开放研究基金资助项目(W200912)

摘　　要：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Pare-to解分布更均匀。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.

关 键 词：多目标优化 微粒群算法 记忆体 多样性 pbest 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]