Spatiotemporal Network Model for Prediction of Silicon Content in Hot Metal of Blast Furnace

In addressing the prediction issue of silicon content in blast furnace hot metal, previous studies often overlooked the analysis of operational mechanisms in the blast furnace. In response, this study integrates mechanistic knowledge to construct a spatial-temporal network structure (K-STN) that reflects the dynamic interactions of various factors during the smelting process. This approach uses a knowledge graph to represent the spatial relationships among variables within the blast furnace, and combines Graph Convolutional Networks (GCN) with self-attention mechanisms to create a spatial-temporal feature extraction module. This method merges mechanistic analysis with data mining, and improves the model's interpretability. Through practical application, this method has demonstrated superior accuracy and stability in predicting the silicon content of hot metal.