A mechanism and data-driven hybrid model for optimising oxygen volume in the main-blowing stage of BOF steelmaking

Accurate control of oxygen volume in the main-blowing stage is essential for basic oxygen furnace (BOF) steelmaking, yet it remains a persistent challenge. This study proposes a hybrid modelling framework that integrates metallurgical mechanisms with machine learning. The proposed framework consists of two sequential stages: determination of optimised oxygen volume and construction of a predictive model. In the first stage, similar heats were selected from the dataset. Based on these selected heats, a decarburisation model for the second-blowing stage was used to calculate the optimised oxygen volume with the aim of achieving a TSC-measured carbon content within the target range. In the second stage, key input features were screened and identified by integrating metallurgical principles with Pearson correlation coefficient analysis. A stacking ensemble model was developed based on the selected input features to predict the optimised oxygen volume. Under the heat-specific feasible-range criterion, the proposed hybrid model achieved a hit rate of 78.71%, outperforming the 74.57% achieved by the original plant control strategy. SHapley Additive exPlanations (SHAP) analysis was further applied to the proposed prediction model to quantify and interpret the contributions of individual input variables to the model predictions. The improved prediction performance and SHAP-based interpretation collectively demonstrate the effectiveness and interpretability of the proposed hybrid framework for oxygen-blowing control in BOF steelmaking.