Meta-relation-based heterogeneous graph neural network with deep reinforcement learning for flexible job shop scheduling

Abstract

The flexible job-shop scheduling problem is a critical challenge in the smart manufacturing industry. Recent methods that combine graph neural networks with deep reinforcement learning have significantly improved scheduling performance by capturing the complex features of jobs and machines. However, existing approaches still exhibit notable limitations: they often neglect the semantic relationships between nodes and fail to fully account for the heterogeneous characteristics of the relations. To address these issues, this study proposes a deep reinforcement learning approach based on a meta-relation-based heterogeneous graph neural network for solving the flexible job-shop scheduling problem. The proposed method explicitly defines distinct meta-relations to encode semantic information among different nodes and employs specially designed graph embedding techniques to capture the heterogeneity of relational structures. Experimental results on publicly available benchmark instances demonstrate that the proposed approach outperforms traditional heuristic algorithms and several state-of-the-art deep reinforcement learning models. Moreover, although the model is trained on small-scale scheduling problems, it exhibits strong generalization capability when applied to large-scale instances.