A refined prediction model of silicon content based on the Kalman Filter

Prediction of silicon content in hot metal is an important task in the control of blast furnace ironmaking process. Due to the complexity of blast furnace ironmaking process, most predictive models may work well under stable conditions, however, when the production is instable, the performance may deteriorate, which means loss of much valuable information contained in the variables. Actually, the residuals of the predictive model consist of two parts, i.e., unmodelled information and noise. In this paper, a TGARCH model (Threshold autoregressive conditional heteroskedasticity model) is used to predict silicon content in hot metal and the residuals are modeled by a Kalman Filter. The Kalman filter is used to separate the unmodeled information from noise and the captured information is then incorporated into the original model. The proposed method was tested on data collected from a medium-sized blast furnace. Simulation results shows that Kalman filter well improve the accuracy of TGARCH model.