Petri Net Graph Neural Network and Intelligent Decision-Making

In response to the frequent changes in real-world production environments, static scheduling methods that generate complete action sequences often require time-consuming rescheduling. A heuristic design approach based on Deep Reinforcement Learning (DRL) addresses this challenge by building a learnable heuristic function, enabling rapid adaptation to changing conditions. The team has proposed a timed Petri Net transfer learning scheduling strategy based on Graph Attention Networks (GAT) and DRL. Through a reachability graph dataset generation algorithm, small-order processing tasks are used to pre-train the graph attention network, significantly improving the exploration efficiency of DRL in its early stages. This approach allows for the timely derivation of effective heuristic functions for scheduling strategies, even in dynamic production environments.