Dynamic prediction of flue gas information driven by converter mouth flame spectral information

Abstract

This study examines converter steelmaking, focusing on real-time aspects of furnace mouth flame spectral radiation and delays in gas analysis to design a predictive system for converter flue gases driven by spectral data. Using a USB2000+ fiber optic spectrometer and gas analyzer, spectral data from the furnace mouth and CO2 and CO flow data from external flue gas were collected. Spectral features were extracted using fast Fourier transform and polynomial fitting methods, and a sample set was constructed. A fully connected neural network algorithm was then employed to predict cumulative flue gas flow. By analyzing changes in spectral and flue gas information, the study investigated delay causes and used cross-correlation to identify temporal discrepancies. Results show that with CO2 and CO prediction errors within ±70 and ±106, the model's root mean square error ranges from 20.54 to 63.43.