基于多级决策的多智能体自动导航车调度系统  

Multiagent AGV Dispatching System Using Multilevel Decision Methods

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系,上海200030

出　　处：《上海交通大学学报》2002年第8期1146-1149,共4页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金资助项目 (60 1743 0 8)

摘　　要：提出基于多级决策和协作学习的方法来建立自动导航车 ( AGV)调度系统中每个 AGV所需要的动态分布式调度策略 .系统中的每一个 AGV都由一个具有两级决策能力的智能体控制 :在选择级 ,智能体采用 Markov对策框架下的强化学习方法 ,以根据其他 AGV当前的子任务建立自己的最有反应子任务 ;在行动级 ,智能体通过强化学习建立优化的动作策略来完成由选择级确定的子任务 .AGV调度仿真结果证明 ,该方法能提高系统的产量 。This research presented a multilevel decision and cooperative learning method to build dynamic and distributed dispatching policies that an AGV needs in an AGV dispatching system. Each AGV is treated as a rational agent in the system, which has two level decisions: on the option level, an agent adopts a reinforcement learning method under the Markov games architecture and makes decision to execute a subtask with the best response to the other AGV's current option; on the action level, an agent learns an optimal policy of actions for achieving its planned option. Finally, an AGV's dispatching simulation testifies that the method will bring more throughput for the system and keep the throughput stable when the ratio of parts arrival changes.

关 键 词：多智能体 自动导航车 调度系统 多级决策 MARKOV对策 强化学习 动态分布式调度策略 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置] TP182[自动化与计算机技术—控制科学与工程]