基于强化学习的自适应在线规划的应用研究  

Application Research on Adaptive On-Line Planning Based on Reinforcement Learning

作　　者：苏世雄 齐金平[1] 

机构地区：[1]兰州交通大学机电技术研究所,甘肃兰州730070

出　　处：《测控技术》2016年第7期124-127,131,共5页Measurement & Control Technology

基　　金："十二五"国家科技支撑计划项目(2012BAH20F05);甘肃省财政厅基本科研业务费(214153);兰州交通大学青年科学基金(2012012)

摘　　要：随着互联网的迅速发展,自适应系统受到越来越多的关注,目前,大部分自适应系统的规划都是预先定义的,但是在开放的互联网环境中,这种预知的规划策略往往灵活性和智能性不高,针对系统运行环境通常是动态的、不确定的,系统设计阶段难以预测到环境所有可能的变化,在设计阶段系统针对环境变化所采取的自适应行为也是难以预先确定的。因此,提出一种系统运行时根据环境的变化在线制定规划的自适应行为策略。采用基于强化学习和Agent技术,对自适应系统的行为进行描述、分析,最后通过相关实验对该策略进行验证,结果表明该在线规划具有自适应能力。With the rapid development of Internet,more and more people attention to the adaptive system,most of the adaptive system planning are defined in advance,but in the open Internet environment,the flexibility and intelligence of this kind of unpredictable planning strategy are not high.In view of that the system running environment is often dynamic and uncertain,the all possible changes of the environment are difficult to be predicted in the system design phase,and the adaptive behavior adopted by the system according to environment changes is difficult to be determined in advance.Therefore,an adaptive online behavior of the planning strategy is proposed based on running system changes in the environment.Based on reinforcement learning and Agent technology,the behavior of the adaptive system is described and analyzed.Finally,validated by relevant experiments of the strategy,the results show the adaptive ability of the online program.

关 键 词：自适应系统 强化学习 在线规划 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]