Prediction of BOF Endpoint P Content by Optuna-RF with K-Medoids Clustering Based on Metallurgical Parameter Constraint

The accurate prediction of the end-point phosphorus content is very important for the control of the BOF steelmaking process. Based on the production data, the random forest model integrating the K-Medoids clustering based on metallurgical parameter constraint, hyperparameter Optuna optimization and interpretability analysis (Optuna-RF) is proposed for predicting the end-point phosphorus content, by comparison with those of the four machine learning models of RF, XGBoost, NGBoost, and LightBGM. The Optuna-RF model uses Optuna's TPE intelligent sampling and to automatically optimize the random forest hyperparameters with 48% importance of the key parameter max_features contribution, reducing the manual workload of parameter tuning. The measurement error is significantly lower than those of the comparison models, with RMSE of 0.001862mass% being 44% lower than traditional RF, and MAE of 0.000969mass% being 83% lower than XGBoost. The Optuna-RF model has the smallest error fluctuation range and the highest peak value in the prediction of endpoint phosphorus content, with a hit rate of 86.55% in the error interval -15, 15; and a hit rate of 97.31% in -25, 25, which has much higher stability and reliability than those of four machine learning models. Characteristic importance analysis of the Optuna-RF model using SHAP analysis reveals the key process parameters affecting the endpoint phosphorus content, which provides a theoretical basis and data support for the optimization of the steelmaking process and process control.