一种多策略融合的多目标粒子群优化算法  

A Multi-Objective Particle Swarm Optimization Algorithm Integrating Multiply Strategies

作　　者：谢承旺[1,2] 邹秀芬[1] 夏学文[2] 王志杰[2] 

机构地区：[1]武汉大学数学与统计学院,湖北武汉430072 [2]华东交通大学软件学院,江西南昌330013

出　　处：《电子学报》2015年第8期1538-1544,共7页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61165004);国家自然科学基金重大研究计划培育项目(No.91230118);江西省自然科学基金(No.20114BAB201025);江西省教育厅科技项目(No.GJJ12307;No.GJJ14373)

摘　　要：为提高多目标粒子群算法在解决复杂多目标优化问题中的整体性能,提出一种多策略融合的多目标粒子群算法.该算法采用均匀化与随机化相结合的方式初始化种群,在粒子速度更新中新增一扰动项,运用简化的k-最近邻方法维持档案以及对档案个体赋予生存期属性并动态调整生存期值.实验结果表明,在GD和SP性能指标上,本文算法与另外5种对等算法在ZDT和DTLZ系列测试问题上进行对比,其表现出了总体显著性的性能优势.In order to improve the overaU performance of multi-objective particle swarm optimization algorithm （MOPSO） in solving complicated multi-objective optimization problems, a multi-objective particle swarm optimization algorithm integrating multiply strategies （MSMOPSO） was proposed in the paper. A new initialization approach of combining uniformization and randomizafion was adopted in the MSMOPSO. Secondly, a disturbance item was added to the particle＇ s velocity updating formula. Thirdly, a simplified k-nearest neighbor approach was applied to preserve the diversity of external archive. Finally, every non-dominated particle in the external archive was assigned the property of lifespan and the lifespan value would be adjusted dynamically during the run of the MSMOPSO. The experimental results illustrate that the proposed algorithm significantly outperforms the other five peer competitors in terms of GD, SP on ZDT and DTLZ test instances set.

关 键 词：粒子群优化 多策略融合 多目标优化问题 多目标粒子群优化算法 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]