【摘要】 本文考虑可数状态空间非平稳马尔可夫决策过程（ＭＤＰ）的平均目标．首先，我们指出并改正了Ｐａｒｋ，ｅｔ，ａｌ［１］和Ａｌｄｅｎ，ｅｔａｌ［２］的错误，并在弱于Ｐａｒｋ,ｅｔａｌ［１］的条件下，借助于新建立的最优方程，证明了最优平均值的收敛性和平均最优马氏策略的存在性．其次，给出了ε（＞０）－平均最优马氏策略的滚动式算法．更多还原

【Abstract】 in this paper, we consider denumerable state nonstationary Markov decisionprocesses with average criterion. First, we point out and correct some mistakes in Park, et al.~[1]and in Alden, et al.~[2]. By the optimal equation built in this paper, we prove the existence ofε(≥ 0)-optimal Markov policies for average criterion and the convergence of optimal averagevalue under conditions which are weaker than those used Park, et al.~[1]. Secondly, we providea rolling horizon algorithm for ε(> 0)-optimal policies.更多还原