摘要
本文考虑由公理化所描述的抽象遗传算法 ,证明了算法种群列以概率 1完全收敛到最优种群集。所获结果应用到具体的遗传算法策略时 ,能明确提出各有关参数的设置策略 ,使之具有所述收敛性 ;当变异概率趋于零时 ,证明了种群列依概率收敛到一致最优种群集。对父代种群参于竞争和杰出者选择遗传算法 。
An axiomatic model of simulated evolutionary computation was introduced recently by the authors (IEEE Trans. Evolutionary Computation, in press in 2001), and, based on the axiomatic model, a general global convergence analysis of genetic algorithms and evolutionary strategy was conducted. This paper continues such study by further relaxing the axiomatation of selection and evolution operators, and providing deeper convergence results of genetic algorithms. Particularly, we present a set of very general conditions which assures the probabilistic and almost-surely probabilistic convergence of genetic algorithms. The formulated conditions are directly dependent of the involved GA parameters, and therefore, provide some useful guidance for the parameter choice in implementation of genetic algorithms.
出处
《工程数学学报》
EI
CSCD
北大核心
2001年第1期1-11,共11页
Chinese Journal of Engineering Mathematics
基金
国家自然科学基金资助项目 !(699750 1 6)
