<p>Monitoring silicon content in molten iron is crucial for indirectly controlling furnace temperatures, optimizing the smelting process, and enhancing product quality. Traditional prediction methods are limited by their inability to capture short-term fluctuations and long-term delayed effects in the smelting process, resulting in poor adaptability and accuracy for complex blast furnace operations. To address these challenges, this paper proposes a novel hybrid deep learning model integrating multi-scale one-dimensional Convolutional Neural Networks (1D-CNN) with Extreme Gradient Boosting (XGBoost). The 1D-CNN architecture efficiently handles temporal data, capturing local features across multiple scales while reducing computational complexity. It precisely identifies critical short-term disturbances and long-term time-lag effects, making it suitable for real-time industrial prediction tasks. Combining 1D-CNN’s feature extraction capabilities with XGBoost’s robust nonlinear regression, the proposed model overcomes existing limitations, achieving accurate and efficient silicon content prediction. Validated with production data, the 1D-CNN-XGBoost model outperforms existing methods, enhancing operational efficiency and stability. This advancement offers a promising solution for optimizing ironmaking processes. Future work may focus on refining the model to handle additional complexities and expanding its application to other industrial process optimizations.</p>