成长性的粒子群算法及其在函数优化中的应用

Particle Swarm Optimization Algorithm with Growth Property and Its Application in Function Optimization

针对传统粒子群优化(PSO)算法在处理复杂函数优化问题时容易陷入局部最优、迭代后期收敛速度慢的问题,提出一种具有成长特性的粒子群优化算法(GPPSO).该算法根据人类成长的特性被分为3个阶段:前期阶段,为速度更新公式增加叛逆项,以降低进入早熟收敛的概率;中期阶段,为平衡全局与局部的搜索,通过对粒子群信息的整合,为速度更新公式添加平衡项;后期阶段,在速度更新公式中去除速度项,充分利用前期粒子进化得到的经验进行局部寻优.同时给出成长阶段划分的两个依据.运用典型函数进行测试,实验表明该算法对于提高收敛性能具有明显优势.

The traditional particle swarm optimization algorithm( PSO) trends to fall into local extremes and has slow convergence rate in the later stages of iteration when dealing with high-dimensional complex functions.To solve this problem,we propose a particle swarm optimization algorithm with growth property( GPPSO).According to the characteristics of human growth,the algorithm is divided into three stages. In the early stage,the speed update formula adds a rebellious item,so that the algorithm can reduce the probability of premature convergence. In the middle stage,in order to make a compromise between the global and local searches,a balance item is added to the speed update formula. In the later stage,the algorithm removes the velocity item in the speed update formula,which is conducive to rapid convergence. Simultaneously,the two bases for the growth stage division are given. GPPSO is applied to some well-known benchmark functions,and experimental results show that it has advantages for improving convergence performance.