基于自适应搜索中心的骨干粒子群算法  

Improved Bare Bones Particle Swarm Optimization with Adaptive Search Center

作　　者：王东风[1] 孟丽[1] 赵文杰[1] 

机构地区：[1]华北电力大学控制与计算机工程学院,河北保定071003

出　　处：《计算机学报》2016年第12期2652-2667,共16页Chinese Journal of Computers

基　　金：高等学校博士学科点专项科研基金(20120036120013);河北省自然科学基金(F2014502059);中央高校基本科研业务费专项资金(2014MS139)资助~~

摘　　要：该文在对标准粒子群算法(Particle Swarm Optimization,PSO)和骨干粒子群算法(Bare Bones Particle Swarm Optimization,BBPSO)中粒子位置的概率密度函数进行分析比较的基础上,对BBPSO进行了改进,并证明了改进算法以概率1收敛于全局最优解.在改进算法中,主要包括如下策略:(1)基于粒子间适应值的差异,提出一种对粒子位置高斯采样均值的自适应调整策略,分析了其作用机理,提出的搜索中心自适应调整策略增加了粒子分布中心的分散度,减缓粒子在中心的聚集趋势;(2)提出了一种"镜像墙"的越界粒子处理方法,该方法能够大幅度地提高算法找到最优解的概率;(3)粒子在不同的进化时期按不同的拓扑结构选取榜样粒子:算法前期主要采用随机结构以增加群体的多样性,算法后期主要采用全局结构以使得搜索更加精细.将该文提出的算法与多种形式的改进PSO,如GPSO(Global PSO)、LPSO(Local PSO)、FIPS(Fully Informed Particle Swarm)、CLPSO(Comprehensive Learning PSO)、HPSO-TVAC(Hierarchical PSO with Time-Varying Acceleration Coefficients)、APSO(Adaptive PSO)、DMS-PSO(Dynamic Multi-Swarm PSO)、OPSO(Orthogonal PSO)、OLPSO(Orthogonal Learning PSO)、ALC-PSO(PSO with an Aging Leader and Challengers)等,以及BBPSO的标准版本和改进版本,如BBJ2(BBPSO with Jumps)、ABPSO(Adaptive BBPSO)、SMA-BBPSO(BBPSO with Scale Matrix Adaptation)等,对CEC2013标准函数进行测试,对实验数据进行非参数检验,结果表明该文改进算法的综合表现要优于其他算法.In particle swarm optimization （PSO） and bare bones particle swarms optimization （BBPSO）, particle positions at every generation can be regarded as random variables. Based on the comparison of probability density distribution functions of the position variables in the two algorithms, this paper proposes an improved bare bones particle swarm optimization algorithm, which is guaranteed to converge to global optimum solution with probability one. There are following strategies in improved algorithm. （1） According to the diversity of individual fitness, the mean of Gaussian distribution of each dimension is controlled adaptively. The action mechanism of which was analyzed. This strategy can increase the dispersal of distribution center and reduce the concentration of particle around the center~ （2） A new boundary condition, which works like a mirror wall, is applied. The method can greatly improve the algorithm to find the optimal solution in probability; （3） In order to balance the exploration and exploitation ability for different periods, particle chooses its exemplar according to different population topologies during evolutionary process. At the beginning of the process, random topology is adopted to increase population diversity. At the later stage of the process, global topology is employed for a more accurate search. The algorithm is compared with other improved variations of PSO, e. g. , GPSO（Global PSO）, LPSO （Local PSO）, FIPS （Fully Informed Particle Swarm）, CLPSO （Comprehensive Learning PSO）, HPSO-TVAC（Hierarchical PSO with Time-Varying Acceleration Coefficients）, APSO （Adaptive PSO）, DMS-PSO （Dynamic Multi-Swarm PSO）, OPSO （Orthogonal PSO）, OLPSO（Orthogonal Learning PSO）, ALC-PSO（PSO with an Aging Leader and Challengers）,and it is compared with the standard version as well as the improved versions of BBPSO, e. g. , BBJ2（BBPSO with Jumps）, ABPSO（Adaptive BBPSO）, SMA-BBPSO（BBPSO with Scale Matrix Adaptation）, and so on. The comparison

关 键 词：粒子群算法 骨干粒子群算法 概率密度 搜索中心 全局收敛 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]