Quantitative Analysis of Multicomponent Converter Gas Using Raman Spectroscopy with Light Intensity Correction

Abstract

Accurate and stable online analysis of converter gas composition is essential for calorific value calculation and efficient energy utilization in the steel industry. However, field disturbances such as laser power fluctuations and optical transmission variations can alter Raman scattering intensity and reduce concentration retrieval accuracy. To address this issue, a system-level strategy combining hardware-based light intensity monitoring and data-driven modeling is proposed. Raman spectra are corrected using synchronously monitored scattered-light intensity and processed by wavelet filtering, adaptive baseline correction, and second-derivative transformation. A LightGBM model is subsequently used to quantify H2, CO2, CO, N2, and CH4. The intensity-corrected model achieves coefficients of determination (R2) ≥ 0.9982 and root-mean-square error (RMSE) values ≤ 0.2493 for all components, with the mean RMSE reduced by 57.5% compared with the uncorrected model. In the laser-power disturbance test, the CH4 prediction error decreases to 0.9%. A six-day field test at a steel plant shows good agreement between the Raman-derived calorific values and those measured by the combustion calorimeter, with a residual standard deviation of 103.11 kcal/m3. These results demonstrate that the proposed method is suitable for robust online analysis of complex industrial gases.