In the billet reheating process during steel rolling, the real-time and accurate prediction of the temperature field is a prerequisite for the dynamic regulation of the heating process, which is crucial for ensuring the quality of billet reheating and reducing the energy consumption of the reheating furnace. The most commonly used finite element thermal field simulation and analysis methods are unable to meet the demand for real-time prediction under dynamic working conditions. Furthermore, traditional machine learning methods struggle to handle irregular data that includes spatial location information, resulting in inaccurate temperature field predictions, which in turn affect the furnace control efficiency and the quality of the product. In light of these limitations, this study proposes a Graph Neural Network (GNN)-based temperature field prediction method for steel rolling reheating furnaces. This method considers the interactions between the nodes within the reheating furnace and constructs a temperature field topology graph to effectively capture these interactions, including heat conduction and convection. Additionally, in the feature fusion and state updating mechanism of the model, we introduce process parameters, such as the air-fuel ratio, as additional inputs to enhance the ability of the GNN to handle complex variables. This enables real-time accurate simulation of the internal temperature distribution of the reheating furnace. The experimental results demonstrate that the model prediction error is controlled within 2.9 % and the response time is maintained within 20 ms, thereby validating the reliability and efficacy of the method in practical applications.
Read full abstract