关键词: 分布式车间, 多智能体, 动态调度, 深度强化学习

Abstract: Aiming at the distributed heterogeneous job shop scheduling problem considering transfer time for dynamic insertion of workpieces,a Multi-Agent Deep reinforcement learning method based on Dueling Double-Deep Q-network(MAD3QN) was proposed with the objective of minimizing the total delay time.This challenge involved two interdependent decision processes:workpiece selection and machine allocation,leading to the creation of two different types of agents,each governed by its own Markov decision process.The state representations,action spaces and reward structures for both the workpiece selection agent and the machine allocation agent were carefully defined to facilitate more efficient decision making.Finally,to validate the effectiveness of the proposed method at different scales,comparisons were made with composite scheduling rules;further evaluations demonstrated the superiority of the proposed approach over heuristic scheduling algorithms and other deep reinforcement learning methods at different scales.

Key words: distributed job shop, multi-agents, dynamic scheduling, deep reinforcement learning