分布模型和关联模型的自适应差分进化算法  

Self-adaptive differential evolution algorithm based on constructing distribution and mapping models

作　　者：周美玲[1] 张胜敏[1] 李征[2] 

机构地区：[1]开封大学软件学院,河南开封475000 [2]河南大学计算机与信息工程学院,河南开封475001

出　　处：《计算机工程与设计》2015年第11期2995-2999,3018,共6页Computer Engineering and Design

基　　金：青年科学基金项目(61402150);河南省科技攻关计划基金项目(122102210507)

摘　　要：自适应差分进化算法通常不区分各维变量所用参数,且未考虑参数之间的关联性,为使参数的调整更好地促进算法性能,提出一种基于分布模型和关联模型的自适应差分进化算法。考虑各维变量之间的差异性,对个体的不同维采用不同的缩放因子,记录一定时期内对种群起促进作用的参数取值,利用主成分分析和分布估计方法建立缩放因子的分布模型;利用记录的缩放因子取值和核支持矢量机建立缩放因子与交叉率参数之间的非线性关联模型,预测交叉率取值。通过6组测试函数进行实验,实验结果表明,该算法比对比算法SaDE更具效力,在t检验指标上优于SaDE的函数数量分别达到了30维时的3组和100维时的5组。Many self-adaptive differential evolution algorithms（SaDEs）employ same parameters for each dimension of variable,and do not construct the relationship between parameters,a self-adaptive differential evolution algorithm based on constructing distribution and mapping models was proposed to further improve the performance of SaDEs.Considering differences among dimensions of variable,different scaling factor values were adopted for each dimension and these beneficial parameter values were recorded in a learning period.Meanwhile,the principal components analysis method and the distribution estimation method were used to construct the distribution model.A nonlinear mapping model was established between scaling factor and crossover rate based on kernel support vector machine.The test results of six benchmarks show that the proposed algorithm works better than the SaDE,and the numbers of benchmarks obtaining better values on 30-dimension and 100-dimension reach three and five in ttest index,respectively.

关 键 词：参数自适应 差分进化 分布模型 关联模型 缩放因子 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]