A Flexible Framework for Scheduling Using Petri Net and Deep Reinforcement Learning

Scheduling represents a fundamental requirement under numerous scenarios in various systems. However, the solution progress of scheduling often imposes limitations on the flexibility of these systems. In this paper, the data-virtual-physical spaces of systems are integrated and fused into a flexible framework for scheduling inspired by Digital Twin. The framework tracks the scheduling state in real-time by utilizing Time Petri Nets and makes decisions on scheduling actions assisted by Deep Reinforcement Learning agents trained according to scheduling tasks. Furthermore, this paper details the specific implementation methods for constructing this framework through an experiment on a wafer sorting system as a case in point. The experimental results show that the framework is capable of attaining efficient and flexible scheduling performance on different scheduling tasks.