支持强化学习的协同进化算法研究  

Research on Coevolutionary Algorithm Supporting Reinforcement Learning

作　　者：王静莲[1] 刘弘[2] 柳新华[1] 李少辉[2] 

机构地区：[1]鲁东大学现代教育技术教学部,烟台264025 [2]山东师范大学信息科学与工程学院,济南250014

出　　处：《系统仿真学报》2009年第4期1037-1039,共3页Journal of System Simulation

基　　金：国家八六三高技术资助项目(2003AA1Z2320);山东省科技攻关计划项目(2005GG4201009)

摘　　要：基于GA的三个主要遗传算子,借鉴生物科学中神经元相互刺激与抑制的量化关系,采用数学模型描述了多个体协同式强化学习的动力学行为;同时对传统GA的进化计算方法进行一定改进,提出了支持强化学习的新的协同进化算法;最后,仿真实验数据表明新算法增强了智能搜索的能力,并加快了种群的收敛速度.Referring from the quantitative relation of neurons cells＇ mutual stimulus and suppression in the physiology and based on the three main genetic operators, the dynamical behavior of the multi-individual cooperative reinforcement learning was modeled. Then the evolutionary computation method of the conventional GA was improved and the novel coevolutionary algorithm supporting the reinforcement learning was proposed. The simulated experiments indicate that the improved algorithm has not only strengthened the capacity of intelligent search but also accelerated the convergence speed of the Population.

关 键 词：遗传算法 强化学习 进化算法 协同 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]