Integrated deep learning-driven multi-stage steam forecasting and scheduling optimization for converter energy...

Area of Science:

• Materials Science
• Chemical Engineering
• Industrial Engineering

Background:

• Converter steam is a crucial secondary energy source in steelmaking, vital for cost reduction.
• Current forecasting methods for converter steam are often single-stage and fail to integrate planning and production data effectively.
• The time-segmented nature of steam generation, with limited planning data and fluctuating production signals, poses forecasting challenges.

Purpose of the Study:

• To develop an integrated two-stage, multi-time-scale forecasting framework for converter steam.
• To bridge the gap between long-horizon planning forecasts and short-horizon production forecasts for improved steam scheduling.
• To enhance the recovery and utilization of converter steam for reduced energy costs in steelmaking.

Main Methods:

• Formulated a time-segmented forecasting workflow linking planning and production stages.
• Employed a hyperparameter-optimized Support Vector Machine (IBKA-SVM) for interval steam output forecasts in the planning stage.
• Utilized a DSC-Transformer model for fine-grained rolling steam-flow prediction during the production stage, dynamically updating the baseline.

Main Results:

• The IBKA-SVM model achieved 92.3% predictive performance in the planning stage.
• The DSC-Transformer model achieved 98.7% predictive performance in the production stage.
• The proposed framework consistently outperformed conventional baseline models in experiments on industrial data.

Conclusions:

• The integrated two-stage framework provides practical, multi-time-scale predictive support for converter steam scheduling.
• The models enhance energy-efficient operation by improving steam recovery and utilization.
• This approach addresses the limitations of single-stage forecasting for time-segmented industrial data.