考虑装配协同的分布式两阶段混合流水车间图强化学习动态调度OA

To address the coordination mismatch between cross-factory production and assembly in the two-stage distributed hybrid flow shop scheduling problem(TSDH-FSSP)under dynamic environments,this paper developed a dynamic scheduling method based on graph reinforcement learning(GRL)with assembly coordination.The method constructed a graph-structured representation integrating operation coupling and shop floor states.It employed a convolutional neural network(CNN)and a dual graph attention network(GAT)to capture intra-factory local topological features and cross-stage global coordination fea-tures.It further applied a proximal policy optimization(PPO)algorithm to achieve end-to-end autonomous decision-making under a composite reward mechanism.Simulation results show that the proposed method outperforms the non-dominated sorting genetic algorithm Ⅱ(NSGA-Ⅱ),the improved whale optimization algorithm(I WO A),and deep reinforcement learning scheduling methods(DRL-GCN and DRL-TF)in terms of makespan and relative percentage deviation(RPD)under dynamic job insertion scenarios.The fixed-rule strategy in the second stage reduced training time by 27.11％while maintaining high solution quality.A real-world bearing manufacturing case verifies the effectiveness and robustness of the proposed method in complex dynamic environments and demonstrates its potential for improving production-assembly coordination in distributed manufacturing systems.