Original Article

PCA-LGBM based prediction of hot metal sulfur content

School of Metallurgical Engineering, Anhui University of Technology, Ma’anshan 243002, Anhui, PR China

^* e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.

Received: 9 April 2026
Accepted: 12 June 2026

Abstract

The sulfur content in blast furnace hot metal is a critical quality indicator that governs the mechanical properties and downstream processing of steel products, and its accurate prediction is essential for optimizing blast furnace operations and controlling desulfurization costs. To address the challenges posed by numerous influencing factors, strong coupling among variables, and potential multicollinearity in hot metal sulfur content, this study proposes a prediction method that integrates Principal Component Analysis (PCA) with LightGBM (LGBM). PCA is first applied to reduce the dimensionality of the original process parameters, and a Bayesian optimization algorithm is then introduced to adaptively tune the hyperparameters of the LGBM model, thereby constructing a PCA-LGBM prediction model. Validation using actual industrial production data demonstrates that the proposed model achieves a coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE) that are significantly superior to those of Random Forest (RF), AdaBoost, and a standalone LGBM model. SHAP analysis further identifies five core features, including blast pressure, and reveals that their influence directions are consistent with the desulfurization mechanism, effectively overcoming the black-box limitation of purely data-driven models. This study provides an effective data-driven modeling strategy for the accurate prediction of hot metal sulfur content in blast furnaces and offers technical support for the intelligent operation and quality control of the blast furnace process.