Deep reinforcement learning for solving efficient and energy-saving flexible job shop scheduling problem with multi-AGV

Abstract

The flexible job shop scheduling problem with multi-automatic guided vehicles (FJSP-AGV) exists widely in the industrial field. To enhance production efficiency and conserve energy, efficient and energy-saving FJSP-AGV is studied. Three optimization tasks: optimizing the makespan, optimizing the total energy consumption (TEC), and simultaneously optimizing the makespan and TEC are solved. To address these challenges, a deep reinforcement learning (DRL) framework is developed. Specifically, in the Markov decision process of the scheduling agent, twelve features are extracted from the shop floor, and sixteen composite scheduling rules are used as the action space. Based on the three optimization tasks, two single-reward functions and a weighted comprehensive reward function are presented. Additionally, the deep Q-network algorithm is used to train the scheduling agent. Comprehensive experiments are conducted on 98 test instances to evaluate the performance of the proposed method. The experiment results demonstrate its effectiveness compared to composite scheduling rules, exact methods, meta-heuristic methods, and other DRL methods.