Intelligent Scheduling for Permutation Flow Shop with Dynamic Job Arrival via Deep Reinforcement Learning

Abstract

The dynamic permutation flow shop scheduling problem (PFSP) is receiving increasing attention in recent years. To provide intelligent scheduling for the dynamic PFSP, we solved the dynamic PFSP with new job arrival using deep reinforcement learning (DRL). The mathematical model is established with the objective of minimizing the total tardiness cost of all jobs arriving at the system. The double deep Q network (DDQN) is adapted to solve the studied problem. A large range of instances is provided to train the DDQN-based scheduling agent. The training curve shows the DDQN-based scheduling agent learned to choose appropriate actions at rescheduling points during the training process. After training, the trained model is saved and used to compare with several well-known dispatching rules on a set of test instances. The comparison results show that our trained scheduling agent performs significantly better than these dispatching rules. Our work can provide intelligent decision-making of scheduling for a flow shop under a dynamic production environment.