Flexible Job Shop Scheduling Problem using graph neural networks and reinforcement learning

The Flexible Job Shop Scheduling Problem (FJSP) is an important research topic in the field of manufacturing. Many studies have used Deep Reinforcement Learning (DRL) to learn Priority Dispatching Rules (PDR) to address the FJSP. However, compared to exact methods, there is still significant room for improvement in the quality of solutions. This paper proposes a new end-to-end DRL framework that utilizes Graph Attention Networks (GAN) to extract relevant information from the disjunctive graph. In this framework, we introduce adaptive weights when calculating attention scores, allowing the model to dynamically adjust the attention scores based on the characteristics of the input data. This helps the model more effectively capture key information within the data. To alleviate the network degradation issue and enhance model performance, the features extracted by the aforementioned model are input into a Residual Connection (RC) module for further deep feature extraction. Finally, our model is validated on generated datasets and public benchmarks, with experimental results indicating that the proposed method outperforms traditional PDR methods and the latest DRL approaches.