A Deep Reinforcement Learning Method for Flexible Job-Shop Scheduling Problem

• Published in: 2024 5th International Conference on Electronic Communication and Artificial Intelligence (ICECAI) pp. 184 - 188
• Main Authors: Shao, Changshun, Yu, Zhenglin, Ding, Hongchang, Cao, Guohua
• Format: Conference Proceeding
• Language: English
• Published: IEEE 31.05.2024
• Subjects: Artificial intelligence; Deep reinforcement learning; Electric potential; Job shop scheduling; Minimization; Optimization; Reinforcement learning flexible job-shop scheduling Maximum completion time
• DOI: 10.1109/ICECAI62591.2024.10675292

This paper focuses on the flexible job shop scheduling problem, with the primary optimization goal being the minimization of the maximum completion time. An innovative end-to-end deep reinforcement learning framework is proposed, based on the deep deterministic policy gradient algorithm. The state space includes six operation features and three machine features. The action space consists of a 3-tuple comprising the job, the operation of the job, and the machine used, catering to both machine selection and operation selection requirements. The reward function is intricately linked to the completion time, accurately reflecting the agent's performance in the scheduling process. Experimental results demonstrate the algorithm's effectiveness in terms of solution quality, showing its ability to find scheduling solutions closer to the global optimal solution.