The Controlling of Longitudinal Cracks Defect Based on BO-XGBoost Model
摘要
The application of machine learning (ML) algorithms in steelmaking processes has enhanced their production efficiency. In this study, feature importance analysis and prediction of longitudinal cracks in continuous casting slabs were conducted based on the XGBoost model. The results show that the BO-XGBoost demonstrates superior overall performance, achieving an accuracy of 0.9673. Feature importance was evaluated based on both Gini coefficient and SHAP values, identifying key features like limestone, calcium carbide, and mold flux as critical for defect prediction. Additionally, parameter estimation was performed for these significant features. The output of qualified slabs without longitudinal cracks is significantly increased when the addition of limestone is 39.302-61.383kg/t, calcium carbide is 0.164–0.610 kg/t, and mold flux 1# is applied. These results will provideclear guidance for optimizing the key features to reduce the longitudinal cracks defect in continuous casting slabs.