差异演化算法的数值模拟研究  

Simulation Study of Differential Evolution

作　　者：袁俊刚[1] 孙治国[1] 曲广吉[1] 

机构地区：[1]中国空间技术研究院总体部,北京100094

出　　处：《系统仿真学报》2007年第20期4646-4648,4784,共4页Journal of System Simulation

摘　　要：差异演化作为一种较新的演化算法,具有较强的寻优能力,但其优化性能受差异演化模式类型及演化控制参数取值的影响非常大。通过一组测试函数的数值模拟研究,给出了演化模式合理选取及演化参数(包括种群大小、交叉概率及缩放因子)合适取值的方法,解决了差异演化算法在应用时面临的一系列问题。此外,还基于演化能力较强的差异演化模式DE/rand/1/exp,提出了一种新的演化模式DE/rand2/1/exp,进一步提高了差异演化效率。Differential evolution （DE）, as a new evolutionary algorithm, is characteristic of strong optimization capability. But its performance is strongly influenced by the variant of differential evolution and the value of each strategy parameter including population size, crossover probability and scale factor. Through a great deal of simulations with a set of test functions, the methods and skills of selecting the DE variant and assigning the strategy parameters values were obtained, resolving serials of problems when using DE. Based on the best variant DE/rand/1/exp, a new variant DE/rand2/1/exp was proposed, and its higher efficiency was showed with simulation.

关 键 词：演化算法 差异演化 数值优化 算法改进 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]