摘要
动态可重构制造系统是实现大批量定制的关键需求之一。将强化学习方法应用于可重构动态调度系统 ,给出了一类典型的可重构多机并行调度决策的问题描述 ,针对该问题建立了基于分布耦合控制的强化学习决策方法。所给方法利用状态分解将原问题转化为一组设备控制器的耦合决策问题 ,以简化单个控制器的求解空间。在局部决策中进一步采用递阶的分层强化学习控制方法 ,使得设备排序调度决策与重构决策分离。重构控制器通过排序调度决策信息及外部设备控制器反馈信息调整设备加工模式 ,达到负荷平衡及优化调度的目的。
Dynamic reconfigurable scheduling is an important component of Mass-Customization manufacturing system. For solving a parallel machine reconfigurable scheduling model, we presented a reinforcement learning (RL) based distributed control scheme. The presented algorithm simplified the state space by converting the original problems into coupling decision processes of several cooperative local controllers. Hierarchical RL based control algorithm was also build to separate the local controller 's scheduling decision and reconfigurable decision. The reconfigurable controller makes decision based on the information of scheduling controller and other local controllers. Simulation results show that the presented algorithm is efficienty and adaptive.
出处
《中国机械工程》
EI
CAS
CSCD
北大核心
2004年第19期1728-1732,共5页
China Mechanical Engineering
关键词
强化学习
动态调度
分布控制
重构调度
reinforcement learing(RL)
dynamic schednling
distributed control
reconfigurable schednling
