Deep Reinforcement Learning based Actor-Critic Framework for Decision-Making Actions in Production Scheduling

Abstract

Smart manufacturers system is faced with the planning and scheduling production challenge to achieve high performance. This research regards flexible job shop scheduling as a sequential decision-making problem and deep reinforcement learning-based Actor-Critic framework is proposed to cope with this problem. The proposed model can extract features from the input data using a convolutional network. In each optimal solution, we regard each operation of a job as a decision that contains information; and each decision as a function of five times in job processing which is classified using information from dispatch rules. For the learning by steps, we run a one-step actor-critic and improve the returns by repeated more steps. We compare different learning rates and discount factors with the generated returns to show the performance of decisions taken by the learning agent. Finally, we experiment proposed model on a case study and benchmark dataset with different values of random seeds to ensure the proposed model framework is more effective over a long time. The results indicate that our proposed model has more effective to solve Flexible job shop scheduling problems with big data set that has more than 100 jobs and 100 machines.