摘要
进化算法的理论研究,如收敛性、时间复杂性研究,是当前的一大热点和难点,有关的理论结果并不多。针对二元进化策略(1+1)ES建立时齐马尔科夫过程模型,利用连续状态马氏过程理论证明了与(1+1)ES相关联的马氏过程在一类连续优化问题中具有指数遍历性,在此基础上证明了(1+1)ES在求解此类优化问题时能以概率1最终找到最优解。所提出的分析方法为进化算法的理论研究提供了一条新思路。
The theoretical investigation to Evolutionary Algorithms,e.g.convergence analysis,runtime analysis,is currently a hot topic,and the related theoretical results are few despite many experimental results.This paper established a homogeneous Markov process model associated with（1＋1）ES.It is proved by means of the theory of Markov process with continuous state that the Markov process associated with（1＋1） ES has exponential ergodicity in a class of continuous optimization problem,hence the（1＋1）ES can finally converge to the optimal solution with probability 1 when solving such a problem.The proposed analytic approach provides a new thought to the theoretical research of evolutionary algorithms.
出处
《计算机科学》
CSCD
北大核心
2011年第7期231-234,共4页
Computer Science
基金
国家自然科学基金(61070033
61003066)
教育部博士点基金(20090172120035)
中央科研业务费专项资金(2009ZM0052)资助
关键词
进化计算
进化策略
收敛性
连续优化
马尔科夫过程
Evolutionary computation
Evolution strategy
Convergence
Continuous optimization
Markov process
