作　　者： ; ; ; ; ; ;

机构地区： 东莞理工学院机械工程学院

出　　处： 《现代制造工程》 2011年第12期45-51,共7页

摘　　要： 构建基于重构和调度二元增强学习架构的调度系统,借助重构增强学习系统的行为来实现资源配置、重构制造单元的功能,借助调度增强学习系统的行为来为各制造单元安排加工任务,实现优化各产品加工路径和加工顺序的功能。重构和调度增强学习系统通过状态转移、行为选择和报酬获取进行联系。提出结合函数泛化器的自适应步长增强学习算法的学习机制和学习步长调整机制,通过实验分析了函数泛化器的性能,验证了该算法解决一类重入型的可重构制造车间调度问题的有效性。 A dual Reinforcement Learning （RL） framework based scheduling system is developed to deal with the scheduling problem on the reentrant reconfigurable manufacturing system. The reconfiguration RL system is constructed to optimize the usage of resources and reconfigure the manufacturing cells. The scheduling RL system is constructed to schedule the tasks for all manu-facturing cells and optimize the processing routes and sequences for all products. The two RL systems are connected through state transition, action selection and reward acquisition. An adaptive RL algorithm combined with function approximation is proposed with the mechanism for learning and adjusting the number of learning steps. Experiments show the performance of the function approximator and validate its effectiveness to minimize the makespan of the proposed scheduling problem.

关 键 词： 调度 增强学习 状态 行为 报酬

领　　域： [机械工程]