Deep Reinforcement Learning for Dynamic Flexible Job-Shop Scheduling with Automated Guided Vehicles

The emergence of automated guided vehicles (AGV) has dramatically facilitated shop floor transportation, making production scheduling challenging. Meanwhile, the arrival of dynamic tasks increases the complexity of production scheduling. Therefore, this study focuses on the dynamic scheduling problem with AGVs under new order arrival. First, the mathematical model of dynamic flexible job-shop scheduling problems with AGVs (DFJSPA) is developed. Then, the DFJSPA is modeled as a Markov Decision Process (MDP), where processing and transportation tasks are considered. Next, we propose a dueling double deep Q network (D3QN) algorithm to optimize the problem. The evaluation results under nine scenarios demonstrate that the D3QN algorithm has less tardiness than the composite scheduling rules, which indicates that the D3QN algorithm can achieve high-efficiency decision-making in dynamic manufacturing systems.