Predicting the Alloying Element Yield in Ladle Furnace Using Stacking Ensemble Learning and SHapley Additive exPlanations Analysis

Accurate prediction of alloying element yield has a significant impact on steel product quality, production costs, and refining efficiency. In this study, the stacking ensemble learning and the SHapley Additive exPlanations (SHAP) analysis are utilized, along with Bayesian optimization, to develop a high-precision and explainable prediction model for the alloying element yield. Different evaluation criterion of model is applied to compare the stacking model with the other existing models. The findings indicate that the stacking model outperforms other models in predicting the alloying element yield, achieving a prediction accuracy of 96.1% within an error range of ±5%. The impact of different variables/base learners on the prediction results and the quantitative influence of individual variables/base learners on the prediction results for each heat are clarified using SHAP analysis. This study contributes to achieving narrow-window control of molten steel composition and enhances the explainability of the model.