基于多种群子空间学习的粒子群优化算法

Particle Swarm Optimization Based on Multi-subgroup and Subspace Learning Strategy

针对标准粒子群(particle swarm optimization,PSO)算法易陷入局部最优、进化后期收敛速度慢和收敛精度低的缺点,提出一种基于多种群子空间学习的粒子群优化算法(MSPSO)。算法将种群分成多个子群,除了传统的种群历史最优粒子和全局最优粒子,还引入分群最优粒子和混合粒子,该混合粒子随机选择各子群最优粒子的相关维度混合而成,增加种群多样性,防止算法陷入局部最优。在种群进化后期,算法对子群最优粒子进行子空间学习,帮助算法逃离局部最优,加快收敛速度。在固定评估次数的情况下,对8种经典的测试函数进行仿真实验,相比较经典知名算法如FIPS、HPSO-TVAC、DMS-PSO、CLPSO、APSO等,MSPSO算法不仅在低维和高维仿真实验中,在逃离局部最优、全局收敛速度和收敛精度上,具有绝对的优势。

tandard particle swarm optimization（PSO）has some shortcomings,such as getting trapped in a local minima,slow convergence velocity and low convergence precision in the late evolutionary. A new improved PSO algorithm which is based on Multi-subgroup and Subspace learning strategy（MSPSO）is proposed in this paper. The algorithm is divided into multiple sub populations. In addition to the traditional optimal particle and global optimal particle,the subgroup optimal particle and mixed particle are introduced. Each dimension of the hybrid particle is the dimension of the optimal particle randomly chosen for each subgroup,in this way,the population diversity is increased and the algorithm is prevented from falling into local optimization. Subspace learning strategy is performed for the subgroup optimal particles in the late of evolution,which helps the algorithm escape from the local optimum and improve the convergence speed. Under keeping the number of function evaluations same,the results which are based on the simulation experiment that using eight benchmark functions show that this article not only has a great advantage in fleeing on the ability of local optima,global convergence speed and solution accuracy than the current well-known and recently improved algorithms such as FIPS,HPSO-TVAC,DMS-PSO,CLPSO,APSO.