About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Energy Technologies and CO2 Management: Resource Efficient Processes
|
| Presentation Title |
Dynamic Prediction Method for Basic Oxygen Furnace Waste Heat Steam Combining Data and Mechanism |
| Author(s) |
longbo wu, zhong zheng, yan hu, chao gao, zhipeng yang, zijun fu, zhenyu gao, yanfei wang |
| On-Site Speaker (Planned) |
longbo wu |
| Abstract Scope |
Waste heat recovery from high-temperature flue gas in Basic Oxygen Furnace (BOF) steelmaking generates steam, enhancing energy efficiency and reducing costs. This study proposes a dynamic prediction method for BOF waste heat steam, integrating data and mechanistic approaches. Industrial data analysis reveals steam recovery differences between duplex and conventional BOF modes. A BOF gas generation mechanistic framework guides correlation-based feature selection and dimensionality reduction. An adaptive deep neural network (DNN) with an error tolerance function replaces the conventional loss functions. Clustering algorithms transform batch steam totals into continuous flow curves using curve similarities to generate flow predictions from the production schedules. Real-time data dynamically update predictions. Simulations show the DNN achieves an overall MAE of 0.95 t across modes, outperforming XGBOOST (MAE 1.80 t) and BPNN (MAE 2.43 t). Steam flow predictions yield an MAE of 15.06 t/h and Rē of 0.94, supporting precise dynamic control and efficient steam utilization. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Iron and Steel, Recycling and Secondary Recovery, Machine Learning |