摘要
针对标准PSO在处理复杂高维优化问题时易出现收敛速度慢、陷入局部最优等问题,提出一种多种群子空间学习的粒子群优化算法(MSPSO).该方法构造了一种新的多子群间信息共享模式,提出子空间学习的概念,并对普通粒子和精英粒子分别进行子空间学习.本文算法简单明确,易于实现,具有很强的稳定性、收敛速度快和较好的全局搜索能力.在固定评估次数的情况下,对常用的19个基准测试函数进行了30维和100维仿真实验,实验结果表明本文算法在收敛速度和求解精度上优于最近提出的几种知名算法(如FIPS、HPSO-TVAC、DMS-PSO、CLPSO、APSO等),特别是在高维问题上优势更加明显.
The standard particle swarm optimization (PSO) may easily behave slow convergence and falling into local minima when solving complex high-dimensional optimization problems. To address this issue, this paper presents a multi-population PSO with sub- space learning ( MSPSO ), which constructs a new modal of information sharing among multiple subpopulations. It proposes a concept of subspace learning,in which the ordinary and elite particles are learned in subspace,respectivley. The proposed approach is simple, clear, and easy to implement. It has strong reliability, fast convergence speed, and good global search ability. Under the fixed number of fitness evaluations, simulation experiments are conducted on 19 benchmark functions with D = 30 and 100. Experimental results demon- strate that the proposed algorithm outperforms some recently proposed famous algorithms ( such as FIPS, HPSO-TVAC, DMS-PSO, CLPSO, APSO, etc. ) in terms of the convergence speed and the accuracy of solutions. Especially for high-dimensional problems, the advantages are highlighted.
出处
《小型微型计算机系统》
CSCD
北大核心
2016年第9期2054-2059,共6页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61261039)资助
江西省自然科学基金项目(20122BAB201043)资助
江西省落地计划项目(KJLD13096)资助
关键词
粒子群优化算法
多种群
子空间
高斯学习
Is: particle swarm optimization
multi-population
subspace
Gaussian learning