Prediction Model for Silicon Content of Hot Metal Based on PSO-TCN

The silicon content of hot metal, which is one of the important indexes characterizing the blast furnace temperature, reflects the blast furnace condition, iron output, and energy consumption intuitively. For the limitation of the gated structure of recurrent neural networks and partial parameters which have long-term, progressive, and lagging effects on the silicon content prediction, a particle swarm optimization-temporal convolutional network (PSO-TCN) model was proposed. Box plots and multiple linear regression were employed for data preprocessing, and the Pearson correlation coefficient method was used to select independent variables for the predictive model. The results indicated that PSO-TCN model could predict the silicon content of hot metal accurately, with the MSE of 0.003, the MAE of 0.044, and the R² of 0.985, which provides guidance for blast furnace operators to control smelting temperature.