Abstract:Job-shop scheduling problem (JSP) is used to determine the processing order of the jobs and is a typical scheduling problem in smart manufacturing. Considering the dynamics and the uncertainties such as machine breakdown and job rework of the job-shop environment, it is essential to flexibly adjust the scheduling strategy according to the current state. Traditional methods can only obtain the optimal solution at the current time and need to rework if the state changes, which leads to high time complexity. To address the issue, this paper proposes a dynamic scheduling method based on deep reinforcement learning (DRL). In the proposed method, we adopt the proximal policy optimization (PPO) to find the optimal policy of the scheduling to deal with the dimension disaster of the state and action space caused by the increase of the problem scale. Compared with the traditional scheduling methods, the experimental results show that the proposed method can not only obtain comparative results but also can realize adaptive and real-time production scheduling.

Dynamic job-shop scheduling in smart manufacturing using deep reinforcement learning