不确定SMDP基于全局优化的鲁棒决策问题  

Robust Decision Problems Based on Global Optimization for Uncertain SMDPs

作　　者：刘春[1] 唐昊[1] 程文娟[1] 

机构地区：[1]合肥工业大学计算机与信息学院,安徽合肥230009

出　　处：《系统仿真学报》2005年第11期2704-2707,共4页Journal of System Simulation

基　　金：国家自然科学基金项目(60404009);安徽省自然科学基金项目(050420303);合肥工业大学中青年科技创新群体计划资助

摘　　要：考虑半马尔可夫决策过程(SMDP)在一些系统参数不确定,且性能函数依赖于这些参数时的鲁棒决策问题。这些参数的不确定性不仅导致等价无穷小生成子的不确定性,也导致性能函数的不确定性。论文针对相关参数的情况,分别采用不同的全局优化算法,即填充函数法和模拟退火算法,进行鲁棒控制策略求解。仿真实例说明,全局优化方法的使用保证了平均准则和折扣准则下的计算结果之间当折扣因子趋近于零时的极限关系成立。The robust decision problems for a class of semi-Markov decision processes （SMCPs） are presented with some uncertain system parameters on which the performance function is dependent. The uncertainty of these parameters will lead to the uncertainty of equivalent infinitesimal generator and also the performance function. Different global optimization methods, such as simulated annealing and filled function approaches, were adopted respectively to derive the optimal robust control policy under the case of dependent parameters. With the application of these global optimization techniques, the obtained simulation results for a numerical example show that an average-cost problem is the limitation of a discounted problem as the discount factor goes to zero.

关 键 词：半马尔可夫决策过程 性能势 鲁棒决策 全局优化 

分 类 号：TP202[自动化与计算机技术—检测技术与自动化装置]