Research on Job Shop Scheduling Rules Based on DQN

To address the issue of scheduling rule constraints on scheduling solutions in job-shop scheduling, a DQN model is established for training. By combining random action selection and experience replay learning, the range of selectable scheduling solutions is expanded. During the training process, scheduling rules are identified and recorded. After training, the scheduling solution and the top 5 most frequently used scheduling rules are output. This enhances the interpretability of the utilization of rules in deep reinforcement learning for job-shop scheduling and provides reference suggestions for the pre-set scheduling rules in the subsequent application of deep reinforcement learning in the job-shop scheduling field. Experiments show that the proposed DQN scheduling model can output relevant scheduling rules in small-scale job-shop scheduling, and the use of scheduling rules meets the requirements of real job-shop scheduling.