Applying Multi-agent Reinforcement Learning and Graph Neural Networks to Flexible Job Shop Scheduling Problem.

The Flexible Job Shop Scheduling Problem (FJSP) is a generalized scheduling problem for modeling production systems. To solve the FJSP, both mathematical optimization and metaheuristic methods have traditionally and widely been widely used. Recently, deep reinforcement learning (DRL) methods have also been employed due to advancements in this field. In particular, the combination of Multi-Agent Reinforcement Learning (MARL) and Graph Neural Networks (GNN) has gained attention as an approach for FJSP. In previous studies that applied MARL and GNN, jobs and machines were defined as agents. However, since the number of jobs is dynamic, there is a scalability issue when the number of job agents increases. To overcome this limitation, this study models the FJSP as a machine pairwise graph structure and applies GNN to reflect cooperation among agents. Furthermore, the study seeks to enhance previous research by applying the distributional value decomposition network (DDN), an state-of-the-art MARL algorithm.