Dynamic flexible job shop co-scheduling optimization based on graph neural network and deep reinforcement learning

With the continuous advancement of manufacturing, job shop scheduling faces complex dynamic disturbances that pose significant challenges to production. This paper investigates the dynamic flexible job shop scheduling problem with job priority and transportation time constraints (DFJSP-PT). The study considers two types of dynamic events: random job arrivals and urgent order insertions implemented through a predefined job priority mechanism. To address the limitations of traditional scheduling methods under complex dynamics, this paper proposes a hybrid scheduling framework based on graph neural network (GNN) and deep reinforcement learning (DRL). The method constructs a Markov Decision Process (MDP) and employs a heterogeneous graph neural network to model the job scheduling state. When new jobs arrive, it enables incremental dynamic expansion of the graph structure, thereby avoiding the need to reconstruct the entire state space. The framework integrates transport time and job urgency into the decision-making process. It dynamically adjusts priorities through a weighting mechanism to achieve joint optimization of operation sequencing and machine allocation. Furthermore, the method introduces a priority experience replay (PER) mechanism based on temporal difference error. This mechanism is combined with composite dispatching rules and a global elite retention strategy, which enhances the algorithm's adaptive learning capability for random job arrivals and emergency order insertion events. Experimental results demonstrate that the proposed algorithm significantly outperforms traditional methods in both convergence performance and solution quality. The algorithm provides an effective technical pathway for intelligent job shop scheduling in dynamic production environments.