求解高维复合体函数的智能优化算法  

Intelligent Optimization Algorithm for High-Dimensional and Complex Functions

作　　者：徐东方[1] 郭战伟[2] XU Dong-fang GUO Zhan-wei(Basic Courses Department, Henan Polytechnic, Zhengzhou 450046, China Huashang College, Guangdong University of Finance and Economics, Guangzhou 511300, China)

机构地区：[1]河南职业技术学院基础教学部,河南郑州450046 [2]广东财经大学华商学院会计学院,广东广州511300

出　　处：《数学的实践与认识》2016年第19期205-211,共7页Mathematics in Practice and Theory

基　　金：国家自然科学基金(61203050);河南省政府决策研究招标课题(2015B213)

摘　　要：由于粒子群算法在处理高维复杂函数时存在容易陷入局部最优的问题,提出了多种群子空间学习粒子群算法(SLPSO),采用多种群进化模式,在粒子更新公式中加入了全局最优粒子,加快了粒子收敛速度,同时在种群之间采用了交叉学习的方法,大大提高了算法的全局搜索能力.另外,还增加了一种子空间学习方法,充分地利用粒子的历史经验,有效地避免了陷入局部最优的问题.通过在高维基准测试函数的仿真实验表明,SLPSO算法的测试结果都明显优于其他两种算法,随着函数维数增加,SLPSO算法测试结果的下降幅度明显低于其他两种算法.在6个极其复杂的复合函数的测试中,SLPSO算法有2个测试函数结果非常接近理论值,其他4个也明显优于其他三种算法.As the problem of the particle swarm optimization algorithm is easy to fall into local optimal for solving high-dimensional and complex functions.A multiple swarms subspace learning particle swarm algorithm（SLPSO） is proposed.The SLPSO algorithm uses multiple population evolution patterns to accelerate the algorithm convergence speed,and the global optimal individual is added in evolution patterns.Meanwhile,the particle of different swarms is cross learning each other to improve the global search ability of the algorithm.What＇s more,the subspace learning method is proposed,making full use of the particle＇s historical experience,also can effectively avoid falling into local optimum.The simulation experiment of high dimension benchmark test functions show that the results of SLPSO algorithm are obviously better than the other two kinds of algorithms.As the dimension of function increased,the results of SLPSO algorithm drop significantly lower than the other two algorithms.In the six extremely complex composition functions tests,SLPSO algorithm has two test function results very close to the theoretical value,and the other four are obviously better than the other three algorithms.

关 键 词：粒子群算法 群体智能算法 高维函数 复合函数 子空间学习 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]