046 Integrating Metallurgical Mechanisms and Explainable Deep Learning Methods to Predict Phosphorus Content in Electri

This study explores the integration of metallurgical mechanisms and explainable deep learning methods to enhance the prediction of phosphorus content during electric arc furnace steelmaking. The objective is to improve the operational efficiency and quality of steel production. A novel approach combining mechanistic models with state-of-the-art machine learning techniques was employed to analyze the complex interactions affecting phosphorus levels. Through a series of experiments, various models, including advanced algorithms like FA-MM-TabNet, were developed and their predictive capabilities were thoroughly assessed. The research found that the proposed method significantly outperformed traditional approaches, achieving higher accuracy in phosphorus prediction while ensuring interpretability of the models through techniques like SHAP (SHapley Additive exPlanations). These findings highlight the potential for coupling metallurgical knowledge with modern artificial intelligence tools to facilitate smarter steelmaking processes. The outcomes of this research not only contribute to the scientific understanding of phosphorus behavior in steel production but also pave the way for practical applications in optimizing industrial steelmaking operations.