APP下载

一种新的混合粒子群优化算法

2022-07-21徐生兵蹇柯夏文杰

软件工程 2022年7期

徐生兵 蹇柯 夏文杰

摘  要:针对粒子群算法在进化后期收敛精度低、收敛速度慢,尤其是高维时候容易早熟等问题,提出了一种新的混合粒子群优化算法。新算法首先设计了一种新的惯性权重,使惯性权重取值在进化初期和后期都较为适中;其次,为了有效抑制粒子陷入局部极值,引入了粒子最优速度和最差适应值的概念,并以此为基础,设计了粒子的一种新的自适应变异方式;最后引入了平均收敛率和最小平均收敛代数两个概念,可以更好地评价和比较本文算法的性能。八个标准测试函数在100 维、200 维进行的数值实验证实,新算法收敛精度高,收敛速度快,且有效预防了早熟现象。

关键词:粒子群优化;惯性权重;早熟;变异

中图分类号:TP183     文献标识码:A

Algorithm of a New Hybrid Particle Swarm Optimization

XU Shengbing, JIAN Ke, XIA Wenjie

(School of Computer and Information, City College of Dongguan, Dongguan 523419, China)

xusb@ccdgut.edu.cn; jianke@ccdgut.edu.cn; xiawj@ccdgut.edu.cn

Abstract: This paper proposes a new hybrid particle swarm optimization algorithm to solve the problems of low convergence accuracy, slow convergence speed of particle swarm optimization algorithm in the late evolution stage, and being prone to mature early especially in the high-dimensional case. The new algorithm first proposes to design a new inertia weight, which makes the value selection of inertia weight moderate in the early and late evolution. Secondly, in order to effectively restrain the particles from falling into the local extreme value, the concepts of particle optimal velocity and the worst fitness of particles are introduced. Based on this, a new adaptive mutation method of particles is designed. Finally, the concepts of average convergence rate minimum average convergence algebra are introduced, which can better evaluate and compare the performance of the proposed algorithm. The numerical experiments of 8 standard test functions in 100 and 200 dimensions verify that the new algorithm has high convergence accuracy, fast convergence speed, and effectively prevents premature phenomenon.

Keywords: particle swarm optimization; inertia weight; premature; mutation

1   引言(Introduction)

粒子群优化(PSO)算法是由Eberhart、Kenned于1995 年提出的一种基于种群智能行为的优化算法。由于其概念明确、需要设置的参数少、编程易于实现等优点,目前在工程领域已经得到广泛的应用。但PSO算法在优化复杂函数问题时极易陷入局部极值,并且会出现早熟现象,这又限制了其进一步应用。为提高PSO算法的优化性能,许多学者提出了各种改进的PSO算法[1-10]。其中,在惯性权重方面,SHI等人[1]于1998 年首次引入了惯性权重的概念,并提出了一种线性递减的权重策略(LDWPSO);黄轩等人[2]提出了一种基于随机惯量权重的快速PSO算法(Faster PSO with Random inertia weight,FRPSO),即惯性权重在0.4—0.6的随机取值;LIANG等人[3]提出了一种利用种群质心和种群个体极值质心的PSO算法(PSO with Centroid,CPSO),使得粒子的更新不仅与传统PSO算法中的个体极值和全局极值相关,而且还与种群中其他粒子的位置和个体极值相关。……

登录APP查看全文