电子学报 ›› 2018, Vol. 46 ›› Issue (8): 1858-1865.DOI: 10.3969/j.issn.0372-2112.2018.08.009

• 学术论文 • 上一篇    下一篇

基于多种群的自适应迁移PSO算法

邓先礼1, 魏波1, 曾辉2, 桂凌3, 夏学文1   

  1. 1. 华东交通大学软件学院, 江西南昌 330013;
    2. 新疆工程学院计算机工程系, 新疆乌鲁木齐 830023;
    3. 华东交通大学经济管理学院, 江西南昌 330013
  • 收稿日期:2017-06-14 修回日期:2018-01-22 出版日期:2018-08-25
    • 通讯作者:
    • 夏学文
    • 作者简介:
    • 邓先礼 男,1978年生,湖北十堰人,硕士,现为华东交通大学软件学院讲师,研究方向为计算智能、复杂网络;魏波 男,1983生,湖北天门人,博士,博士后,现为华东交通大学软件学院讲师,研究方向为计算智能、机器学习;曾辉 男,1980年生,湖北麻城人,博士,现为新疆工程学院计算机工程系副教授,研究方向为计算智能;桂凌 女,1977年生,湖北应城人,学士,现为华东交通大学经济管理学院实验师,研究方向为智能物流.
    • 基金资助:
    • 国家自然科学基金 (No.61663009,No.61602174,No.61763010,No.61762036); 江西省交通厅科技项目 (No.2017D0038); 江西省教育厅自然科学基金 (No.GJJ160469); 新疆维吾尔自治区教育厅高校科研计划基金 (No.2014JYT041606); 新疆工程学院博士科研启动基金 (No.2015BQJ011712)

A Multi-Population Based Self-Adaptive Migration PSO

DENG Xian-li1, WEI Bo1, ZENG Hui2, GUI Ling3, XIA Xue-wen1   

  1. 1. School of Software, East China Jiaotong University, Nanchang, Jiangxi 330013, China;
    2. Department of Computer Engineering, Xinjiang Institute of Engineering, Urumqi, Xinjiang 830023, China;
    3. School of Economics and Mangement, East China Jiaotong University, Nanchang, Jiangxi 330013, China
  • Received:2017-06-14 Revised:2018-01-22 Online:2018-08-25 Published:2018-08-25

摘要: 针对标准PSO中单一社会学习模式造成的算法容易陷入局部最优和后期收敛速度慢等问题,提出了一种基于多种群的自适应迁移PSO算法(Multi-population based self-adaptive migration PSO,MSMPSO).通过融合两种常用的邻居拓扑结构,赋予个体更多的信息来源;在多个子种群并行进化的基础上,利用不同加速因子的组合赋予各子种群不同的搜索特性,进而通过周期性对子种群的历史性能进行评估,以此为基础指导个体的迁移操作,实现子种群间的协作与计算资源的合理分配,并最终提升算法的综合性能.对CEC2013测试函数的优化结果表明,MSMPSO在求解精度、收敛速度等方面均表现出较好的性能.

关键词: 粒子群算法, 社会学习, 多种群, 个体迁移, 历史性能评估

Abstract: The performance of particles' "social-learning" ability directly affects the search capability of PSO.To overcome some shortcomings caused by mono-social-learning model,such as premature convergence and slow convergence speed at later evolution stage,a multi-population based self-adaptive migration PSO (MSMPSO) is proposed.In MSMPSO,the two common neighbor typologies are integrated into particle's social-learning part aiming to give more information source for the particles.Furthermore,the entire population is divided into three sub-populations which are evolved in parallel.Based on the multi-population mechanism,different search characteristics caused by three different combinations of acceleration coefficients are assigned to the three sub-populations.To take advantage of different merits of different sub-populations,and realize the reasonable allocation of computing resources,individuals carry out a migration operator based on subpopulations' historical performance during the last period.Simulation results based on CEC2013 test suite manifest that the favorable comprehensive performance of MSMPSO,in terms of convergence speed and solutions accuracy.

Key words: particle swarm optimization, social learning, multi-population, individual migration, historical performance evaluation

中图分类号: