Markov决策过程不确定策略特征模式  

Property Patterns of Markov Decision Process Nondeterministic Choice Scheduler

作　　者：黄镇谨[1,2] 陆阳[1,3] 杨娟[1] 方欢[1] 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009 [2]广西工学院计算机工程系,柳州545006 [3]安徽省矿山物联网与安全监控技术重点实验室,合肥230088

出　　处：《计算机科学》2013年第4期263-266,共4页Computer Science

基　　金：国家自然科学基金资助项目(60873195;61070220);高等学校博士点基金资助项目(20090111110002)资助

摘　　要：马尔科夫决策过程可以建模具有不确定性特征的复杂系统,而在进行模型分析时需要采用策略对不确定性进行处理。首先,研究不同策略下时空有界可达概率问题,给出不确定性解决策略的定义及分类方法。其次,在时间无关策略下,证明基于确定性选取动作和随机选取动作的时空有界可达概率的一致性,并且论证了时间依赖策略相对于时间无关策略具有更好的时空有界可达概率。最后结合实例简要阐述了结论的正确性。Markov decision process can model complex system with nondeterminism. Schedulers are required to resolve the nonderministic choices during model analysis. This paper introduced the time-and space-bounded teachability proba- bilities of markov decision process under different schedulers. Firstly, the formal definition and classification method of schedulers for nonderminism were proposed and then we proved that the reachability probabilities coincide for determini- stic and randomized schedulers under time-abstract. Also, it was proved that time-dependent scheduler generally induces probability bounds that exceed those of the corresponding time-abstract. At the end of paper, two cases were illustrated for describing the correctness of the conclusion.

关 键 词：马尔科夫决策过程 不确定性策略 时空有界可达概率 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]