无界非时齐马氏决策规划的矩最优化问题  

A MDP Model of Moments Problem with Unbounded Rewards and Dynamic Transition Probabilities Considering an Initial State

作　　者：段国圣[1] 

机构地区：[1]江汉石油学院基础课部

出　　处：《江汉石油学院学报》1990年第3期78-87,共10页Journal of Jianghan Petroleum Institute

摘　　要：文章讨论无界非时齐马氏决策规划(MDP)的矩最优问题,并考虑初始状态的影响,证明了寻求最优策略等价于寻求最优马氏策略;指出一个k阶矩最优问题可以转化为一个同类型的一阶矩最优问题;给出了存在k阶矩最优策略的充要条件;同时还讨论了最优策略的若干性质。诸如石油工业中勘探方案选择,考虑钻井成本的最优化等复杂问题,可以归结为非线性随机动态规划问题,MDP方法是解决这些问题的良好途径。A MDP(Markov Decision Programming) model of moments problem with unbounded rewards and dynamic transition probabilities considering an initial state is dealt with in this paper It is shown that a k-th moment problem can be transformed into a first moment problem. A necessary and sufficient condition for π~* to be a k-th moment optimal policy is established. Finally, the properties of an optimal policy are investigated. The MDP model discussed above may point out an efficient way of solving the problems of nonlinear, random and dynamic progamming nature in selecting exploration alternatives and optimizing drilling cost in the oil industry.

关 键 词：钻井 成本 无界报酬 MDP 决策 

分 类 号：TE22[石油与天然气工程—油气井工程]