<p>Accurately predicting oxygen consumption in basic oxygen furnace (BOF) steelmaking is essential for improving efficiency and quality. However, existing methods often face challenges such as data noise, nonlinear relationships, and limited interpretability. To address these issues, a hybrid model integrating mechanistic and data-driven approaches is proposed. The Oxygen Balance Mechanism (OBM) serves as the foundation, providing physics-based interpretability and generating intermediate outputs for the Hybrid Oxygen Balance Mechanism (HyOBM). BP neural network (BP) is employed to capture complex nonlinear relationships and boost prediction accuracy, while Gaussian Process Regression (GP) is employed to suppress data noise and provide uncertainty estimates. The outputs of HyOBM, BP, and GP are fused through weighted averaging to construct the final model. By combining the interpretability of OBM with the nonlinear modeling capacity of BP and the noise suppression of GP, the model achieves a balance between interpretability, accuracy, and robustness, effectively addressing challenges in industrial oxygen consumption prediction. The proposed model outperformed other HyOBM-based models, delivering substantial improvements in both prediction accuracy and reliability.</p>