Dynamic Prediction of Reblowing Necessity in BOF Steelmaking

The basic oxygen furnace (BOF) steelmaking process is the primary route for steel production but faces challenges in efficiency, emissions, and precise endpoint control. This study introduces a novel approach to dynamically predict the neces-sity of reblowing using time-series data from sensors, additive patterns, and static variables collected over a year of industrial production. By integrating time-series data and extracting statistical and frequency-domain features, we developed machine learning models, incorporating various sampling techniques to address class imbalance. We applied window encoding to process time-series information, enabling predictions from the start of the process through to the first sampling stage. The XGBoost model, coupled with synthetic sampling, demonstrated the highest performance. Our approach classifies heats as normal or abnormal based on the need for reblowing, providing valuable insights for BOF process control. The continuous improvement in prediction accuracy throughout the process highlights the effectiveness of this method, marking the first study to dynamically predict reblowing necessity in the BOF process and emphasizing the importance of process data in optimizing BOF operations.