Optimization of Dynamic Scheduling for Flexible Job Shops Using Multi-Agent Deep Reinforcement Learning

This study proposes an optimization framework based on Multi-agent Deep Reinforcement Learning (MADRL), conducting a systematic exploration of FJSP under dynamic scenarios. The research analyzes the impact of two types of dynamic disturbance events—machine failures and order insertions—on the Dynamic Flexible Job Shop Scheduling Problem (DFJSP). Furthermore, it integrates process selection agents and machine selection agents to devise solutions for handling dynamic events. Experimental results demonstrate that, when solving standard benchmark problems, the proposed multi-objective DFJSP scheduling method, based on the 3DQN algorithm and incorporating an event-triggered rescheduling strategy, effectively mitigates disruptions caused by dynamic events.