一种基于混合概率选择算子的改进遗传算法  

Improving Genetic Algorithm based on A Hybrid Probabilistic Selector

作　　者：赵鑫宁 喻歆[1] 吴锡[1] 

机构地区：[1]成都信息工程大学计算机学院,四川成都610225

出　　处：《成都信息工程大学学报》2016年第3期247-254,共8页Journal of Chengdu University of Information Technology

基　　金：国家重点基础研究发展计划(973计划)资助项目(2014CB360500;2014CB360506)

摘　　要：遗传算法(genetic algorithm,GA)已被成功应用于求解实值优化问题,但其在求解多峰实值优化问题时面临收敛较慢和早熟收敛的问题。为解决该问题,提出一种基于高斯分布和柯西分布的概率选择算子。算子在执行选择操作时,分别根据当前种群生成高斯和柯西分布,通过对分布采样获得参加遗传操作的个体,从而在保证选择压力的同时保持一定程度的种群多样性,避免早熟收敛。此外,基于种群算法投资组合(population-based algorithm portfolios,PAP)策略,同时利用2种概率选择算子的优势,采用2个子种群并行搜索,提高算法在多种不同问题上的求解性能。在一系列多峰实值优化问题上的测试结果表明,文中提出的改进方法能够显著提高遗传算法在收敛速度和求解精度两方面的性能。Genetic Algorithms（ GA） have been applied to numerical optimization problems successfully. However,they are confronted with slow convergence and prematurity. Therefore,this paper proposes a hybrid probabilistic selector for a real-encoded GA based on Gaussian and Cauchy distributions. During the selection operation,a Gaussian and a Cauchy distribution are generated based on the selected groups of current population respectively. Then candidate individuals for genetic operations are generated by sampling the two distributions respectively. With the proposed selector,the selection pressure is guaranteed while a certain diversity is maintained which is beneficial to avoidance of prematurity. Furthermore,in order to improve the overall performance of GAs on a wide range of problems,two sub-populations,equipped with Gaussian and Cauchy based selectors respectively,are run parallel by adopting the strategy of PopulationBased Algorithm Portfolios（ PAP）. Experimental results on a set of numerical optimization problems show that the proposed approach can significantly increase the performance of GAs with respect to the convergence rate and the solution quality.

关 键 词：计算机应用 人工智能 遗传算法 多峰函数 高斯分布 柯西分布 种群算法投资组合 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]