Abstract: According to the actual production data of 120 t converter of Zhongtian Iron and Steel Group Co., Ltd., a prediction model of phosphorus content in molten steel at the end of converter blowing based on Stacking algorithm was established. The main factors influencing dephosphorization were determined through the thermodynamic analysis of dephosphorization, and then the input variables of the model were determined. After the data preprocessing was completed, six machine learning algorithms (RF, ET, XGBoost, LightGBM, CatBoost, and NN) were used to establish the models, and then the prediction results of these six models were used for Stacking ensemble modeling by multiple linear regression algorithm. By comparing the prediction results of these seven models, it could be concluded that the Stacking ensemble model had the best prediction effect, and the hit rates of the predicted endpoint phosphorus mass fraction were 90.59 % and 97.56 % when the error interval was ±0.004 % and ±0.005 %, respectively.

Key words: converter steelmaking, prediction of endpoint phosphorus content, ensemble learning, Stacking ensemble