第5卷第5期 智能系统学报 Vol.5 No.5 2010年10月 CAAI Transactions on Intelligent Systems 0ct.2010 doi:10.3969/i.isgn.1673-4785.2010.05.001 多目标微粒群优化算法综述 王艳12,曾建潮2 (1.兰州理工大学电信工程学院,甘肃兰州730050:2.太原科技大学复杂系统和智能计算实验室,山西太原030024) 摘要:作为一种有效的多目标优化工具,微粒群优化(S0)算法已经得到广泛研究与认可.首先对多目标优化问题 进行了形式化描述,介绍了微粒群优化算法与遗传算法的区别,并将多目标微粒群优化算法(OPS0)分为以下几 类:聚集函数法、基于目标函数排序法、子群法、基于Pto支配算法和其他方法,分析了各类算法的主要思想、特点 及其代表性算法.其次,针对非支配解的选择、外部档案集的修剪、解集多样性的保持以及微粒个体历史最优解和群 体最优解的选取等热点问题进行了论述,并在此基础上对各类典型算法进行了比较.最后,根据当前MOS0算法的 研究状况,提出了该领域的发展方向. 关键词:多目标优化;微粒群优化算法:非支配解:外部档案:多样性 中图分类号:1P18文献标识码:A文章编号:16734785(2010)050037708 A survey of a multi-objective particle swarm optimization algorithm WANG Yan12,ZENG Jian-chao2 (1.College of Electrical and Information Engineering,Lanzhou University of Technology,Lanzhou 730050,China;2.Complex System and Computational Intelligence Laboratory,Taiyuan University of Science and Technology,Taiyuan 030024,China) Abstract:Particle swarm optimization(PSO)algorithms have been widely studied and approved as effective multi- objective paper optimizers.In this paper,first of all multi-objective problems were formally described,and the difference between a PSO and genetic algorithm (GA)was introduced.Then the taxonomy of current multi-objec- tive PSO(MOPSO)algorithms,which include aggregate functions,sorting based on objective functions,sub-popu- lation methods,Pareto dominated based algorithms,and other algorithms,was presented.Additionally,the main i- deas,features,and representative algorithms of each approach were analyzed.Secondly,hot topics in MOPSO al- gorithms such as selecting non-dominated solutions,pruning archive sets,maintaining the diversity of the solutions set,and selecting both the best personal and global solutions were discussed on the basis of which all typical algo- rithms were compared.Finally,several viewpoints for the future research of MOPSO were proposed according to the present studies. Keywords:multi-objective optimization;particle swarm optimization;non-dominated solutions;archive;diversity 在科学实践、工程系统设计及社会生产活动中,知道每个目标函数所占的权重,并且对目标给定的 许多问题都是多目标优化问题.通常多目标优化问次序也比较敏感. 题中的各个目标函数之间可能会存在冲突,这就意 微粒群优化(PS0)算法是1995年由Kennedy 味着多目标优化问题不存在惟一的全局最优解,使 和Eberhart提出的一种基于群体智能的优化算法, 得所有目标函数同时达到最优.为了达到总目标的 应用于单目标优化问题时表现出了快速收敛的特 最优化,需要对相互冲突的目标进行综合考虑,对各 点.随着对PS0研究的深入,该算法已经由用来解 子目标进行折衷.最初,多目标优化问题往往通过加 决单目标优化问题逐步拓展到用来解决多目标优化 权等方式转化为单目标优化问题,但这样需要事先 问题.1999年Moore和Chapman首次提出将PS0算 法应用于解决多目标优化问题,但这个思想未公 收稿日期:20090922. 基金项目:国家自然科学基金资助项目(60674104) 开发表.从此以后用PS0解决多目标优化问题开始 通信作者:曾建潮.E-mail:z心ngjianchao@263.net. 得到研究人员的关注,但直到2002年Coello等21和