Blast furnace conditions evaluation and prediction system based on temporal convolutional network

Precise charging distribution is a core challenge in intelligent ironmaking. However, most existing blast furnace condition prediction methods rely solely on the historical condition and ignore control variables, thus failing to guide charging operations directly. To address this issue, a blast furnace condition time-series evaluation and prediction system is developed using historical charging matrices and condition data. To begin with, a historical database of blast furnace condition is built and a nine-level classification method is proposed. Subsequently, a temporal convolutional network based evaluation model is established to evaluate blast furnace condition time series. Furthermore, a multi-dimensional temporal convolutional network with a dual-branch architecture and additive fusion is designed, which separately extracts features from the historical condition sequence and the charging matrix to predict future blast furnace conditions. Finally, five commonly used charging modes are selected for regulating blast furnace conditions, and the most suitable one is suggested based on the evaluation results. Simulation results indicate that the proposed method can effectively evaluate and predict blast furnace conditions, providing predictive guidance for charging distribution decisions and suggesting potential for improving product quality and safety.