Hot metal temperature forecasting at steel plant using multivariate adaptive regression splines

Abstract

Steelmaking has been experiencing continuous challenges and advances concerning process methods and control models. Integrated steelmaking begins with the hot metal, a crude liquid iron that is produced in the blast furnace (BF). The hot metal is then pre-treated and transferred to the basic lined oxygen furnace (BOF) for refining, experiencing a non-easily predictable temperature drop along the BF–BOF route. Hot metal temperature forecasting at the BOF is critical for the environment, productivity, and cost. An improved multivariate adaptive regression splines (MARS) model is proposed for hot metal temperature forecasting. Selected process variables and past temperature measurements are used as predictors. A moving window approach for the training dataset is chosen to avoid the need for periodic re-tuning of the model. There is no precedent for the application of MARS techniques to BOF steelmaking and a comparable temperature forecasting model of the BF–BOF interface has not been published yet. The model was trained, tested, and validated using a plant process dataset with 12,195 registers, covering one production year. The mean absolute error of predictions is 11.2 °C, which significantly improves those of previous modelling attempts. Moreover, model training and prediction are fast enough for a reliable on-line process control.