基于递阶强化学习的多智能体AGV调度系统  

Multiagent AGV dispatching system based on hierarchical reinforcement learning

作　　者：李晓萌[1] 杨煜普[1] 许晓鸣[1] 

机构地区：[1]上海交通大学自动化研究所,上海200030

出　　处：《控制与决策》2002年第3期292-296,共5页Control and Decision

摘　　要：递阶强化学习是解决状态空间庞大的复杂系统智能体决策的有效方法。具有离散动态特性的AGV调度系统需要实时动态的调度方法 ,而具有 Max Q递阶强化学习能力的多智能体通过高效的强化学习方法和协作 ,可以实现 AGV的实时调度。Hierarchical reinforcement learning is an effective method of solving decision problems for complex systems with enormous number of states. AGV dispatching system needs dynamic dispatching rules because of its discrete and dynamic properties. Multiagent with the capacity of Max Q hierarchical reinforcement learning is implemented in real time AGV dispatching by high performance learning and cooperation. The simulation testifies the efficiency of this method.

关 键 词：递阶强化学习 多智能体 AGV调度系统 机器学习 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]