作　　者： ; ; ; ;

机构地区： 广东工业大学

出　　处： 《中国机械工程》 2009年第12期1439-1445,共7页

摘　　要： 通过分析模具制造项目工期、费用与报酬的不确定性以及模具返修、随机插单等突发事件频繁发生的特点,构建了基于离散时间马尔可夫链的模具制造项目群随机演化模型,提出了求解该随机动态规划模型的算法。通过并行启发式策略进行仿真和多步Q学习,有效解决了"维数灾难"问题,结合示例阐述了算法执行过程,说明了其可行性与可靠性。与标准Q学习方法相比,多步Q学习方法能够有效提高算法收敛速度。 Through the analysis of uncertainties of the durations, costs, rewards as well as the characteristics of frequent repair in mould and die manufacturing projects with new potential projects randomly emerging, this paper proposed a stochastic evolution model of multiple mould and die manufacturing projects, which was on the basis of a discrete time Markov chain. With the aim to overcome the curse of dimensionality~ ,an algorithm combining parallel simulation of heuristic policies and multi--step Q learning was put forward to solve the above stochastic dynamic programming model. Finally, such an algorithm was explained with a sample example. The results show that the model is applicable and the algorithm is reliable and effective as well. With multi--step Q learning method, convergence speed can be effectively improved in the above algorithm compared with standard Q learning method.

关 键 词： 模具制造项目群 随机动态规划 马尔可夫链 并行启发式仿真 多步 学习

领　　域： [机械工程]