Job Shop Scheduling for Dynamic Environments Using a Q-Learning Approach

Efficient scheduling in dynamic production environments with frequent disruptions presents a significant challenge, particularly in the context of the Job Shop Scheduling Problem (JSSP). Traditional solutions, constrained by static assumptions, often fail to adapt to such dynamic environments. To address this issue, this study proposed a Q-learning-based framework for the JSSP, utilizing the adaptability of Reinforcement Learning (RL) to dynamically select scheduling strategies. The numerical experiment using Taillard's JSSP dataset demonstrated that the proposed method achieves a mean makespan of $1594.70 \pm 81.23$, outperforming conventional dispatching rules such as First In, First Out (FIFO), Longest Processing Time (LPT), Shortest Processing Time (SPT), and Most Remaining Time (MRT). It also surpasses the Random Strategy (RS), which yields significantly higher makespan values. These findings illustrated the potential of the Q-learning-based method as a robust alternative to traditional heuristic methods, paving the way for advancements in real-time production scheduling strategies.