基于投影映射的动态多目标粒子群优化算法  

Dynamic Multi-Objective Particle Swarm Optimization Based on Projection Mapping

作　　者：陈兵华[1] 尤嘉兴 陈基漓[1] 董明刚[1] 

机构地区：[1]桂林理工大学信息科学与工程学院,广西桂林541004 [2]厦门市酷游网络科技有限公司,福建厦门361003

出　　处：《计算机仿真》2016年第12期233-238,423,共7页Computer Simulation

基　　金：国家自然科学基金项目(61203109;61563012);广西高校科研项目(KY2015LX112);广西研究生教育创新计划资助项目(YCSZ2014157)

摘　　要：为保证在动态环境中及时跟踪到最新的真实Pareto前沿,针对基于拥挤距离的档案维护策略时间复杂度过高的问题,提出一种采用投影映射的动态多目标优化算法。上述算法利用投影映射的密度估计方法较精准快速地估算档案中解的密度;采用了ε-支配弱化传统的Pareto占优以使非劣解达到均匀分布;引入了对新环境下Pareto解的预测,加快了算法的收敛进程;提出了一种种群-领导集-档案的三层结构,使得算法在动态环境下的多目标优化中表现良好。实验结果表明,改进算法能适应动态环境,快速跟踪动态Pareto面,且解集收敛性及均匀性良好,有效降低了档案维护的时间复杂度。In order to track the latest true Pareto Front timely in dynamic environment, for the high time complex- ity in archive maintenance strategy based on crowding distance, a dynamic multi - objective particle swarm optimization was presented, which and accurately and quickly estimate density of solutions in archives through projection mapping. This algorithm performs well in the dynamic process of multi - objective optimization, as results from the following methods--using epsilon domination to weaken traditional Pareto dominance to make non - inferior solutions evenly - distributed; predicting Pareto solutions in new environment ; accelerating convergence process ; bringing forward the ＂population - leadership set - archive＂ structure. Results manifest that this algorithm can adapt itself to dynamic environment, track dynamic Pareto fast and keep its solution sets in good uniformity and astringency as well as reduce time complexity.

关 键 词：投影映射 动态多目标优化 粒子群 种群-领导集-档案 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]