高维金字塔型协同粒子群算法  

A Pyramid-type Cooperative Approach to Particle Swarm Optimizations with Multi-dimensions

作　　者：张行[1] 刑志栋[1] 董建民[1] 

机构地区：[1]西北大学数学系,陕西西安710069

出　　处：《计算机与现代化》2009年第12期18-20,24,共4页Computer and Modernization

基　　金：陕西省教育厅专项基金资助项目(05JK303)

摘　　要：针对一般粒子群PSO求解高维优化往往陷入局部收敛的"诅咒"问题,设计了一种高维金字塔型协同粒子群优化(A Pyramid-type Cooperative Approach to Particle Swarm Optimizations with Multi-dimensions,PCPSO-M)算法。PCPSO-M算法结合了PSO较快收敛以及CPSO协同算法局部寻优能力强的特点。把粒子群分为三层金字塔型,各个层间、层内相互协同作用,在最上层由于维数过大,则采用一半"较好"适应值的粒子代替另一半"较差"适应值的粒子。这种上下协同,内外"学习"的方法,很好地解决了维数高的问题,弥补了CPSO的不足;尤其在Rosenbrock、Quadric函数的测试中,实验结果表明,解的质量好,效果满意。General particle swarm optimization for solving high-dimensional optimization problems potentially getting trapped in sub-optimal ＂curse＂, the paper designs a pyramid-type cooperative approach to particle swarm optimizations with multi-dimensions algorithm（PCPSO-M）, which combines PSO＇ s earlier convergence with CPSO＇ s stronger searching optimal ability. The particle swarm is divided into three layers, particles of which interacts internally and externally. For excessive dimensions on the top, haft of the ＂good＂ fitness of the particles replace the other half of ＂poor＂ in the fitness particles. So the ＂learning＂ way of coordination is a very effective solution to the problem of high dimension, which makes up for deficiencies of CPSO algorithm ; especiaUy in the Rosenbrock, Quadric function tests, a satisfactory solution can be got.

关 键 词：协同粒子群 粒子群算法 协同学习 金字塔 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]