Prediction of end-point phosphorus content of molten steel in BOF with machine learning models

Abstract

The main task of basic oxygen furnace (BOF) steelmaking is dephosphorization, thus the prediction and control of End-point phosphorus content of molten steel is of great significance. Four machine learning regression models (Lasso, Random Forest, Xgboost, and Neural Network) were established to predict end-point phosphorus content of molten steel in BOF according to raw and auxiliary material data, process parameters, and data of production quality. The prediction effect of four models were further compared, and their prediction results were interpreted via model?s interpretability and Permutation Importance Method. Results showed that compared with linear regression and Neural Network regression model, two kinds of ensemble tree model had higher prediction accuracy, better stability in small data sets, and lower requirements on data pre-processing. The influencing factors of end-point phosphorus (P) content in BOF were ranked by importance as: Tapping temperature > Turning down times > Steel scrap amount > Operation habits of different work groups > Blowing oxygen amount > Sulfur and Phosphorus content of molten iron > Addition amount of lime, limestone, and light-burned dolomite in slagging agents > Slag-splashing amount.