Separation and analysis method of overlapping absorption spectra with cross interference in gas mixture measurement

Abstract

The interference between overlapping gas absorption lines often occurs in the measurement of multi-component gas mixture with using tunable diode laser absorption spectroscopy (TDLAS). This is also the main problem of the technology in some applications. For instance, in the early application of multi-component gas mixture measurement in coal mines, we found that the absorption lines of carbon monoxide (CO) and methane (CH₄) seriously overlapped. The absorption signal of trace CO gas was annihilated and could not be effectively demodulated, especially in the presence of high concentration of CH₄. This problem could not be solved just by accurately selecting the spectral lines due to the band absorption of CH₄. Therefore, in this paper, we introduce the support vector regression (SVR) model to deal with the interference between CO and CH₄ absorption lines. The spectral signals of 14 groups of mixed gases with different concentrations of CO and CH₄ are used as the training sets, and the five-fold cross-validation is adopted to prevent the model from overfitting. After 15 iterations in 30 seconds, the optimal regression model of CO and CH₄ can be obtained respectively. Furthermore, it is worth noting that based on the experimental data, the linear kernel function is selected to construct the two gas SVR models, and the parameters of the SVR models are optimized by the sequential minimal optimization(SMO) algorithm. With the assistance of the SVR models, the absorption spectra of the two gases can be demodulated effectively, and finally the accurate measurement results are obtained. The measurement results show that the absolute error of trace CO and CH₄ concentration(volume fraction of gas) are less than 2 × 10^–6 and 0.2 × 10^–2 respectively. Meanwhile, the correlation coefficient between the measured values and the actual values of CO and CH₄ are 0.998 and 0.9995, respectively. In addition, the dynamic stability for each of the two regression models is fully verified by the experiment of the inflation process. Consequently, this method can eliminate the interference between the overlapping spectra, and can fully meet the requirements for accurately measuring the gas mixture. We hope that the SVR model can provide an effective solution for the real-time monitoring of multi-component gas mixture, and thus greatly improving the adaptability of TDLAS technology in the future.