基于傅里叶变换红外光谱技术的混合毒性气体定量分析研究

Study on the quantitative analysis of multi-component toxic gases based on FTIR

基于傅里叶变换红外光谱技术,选取低浓度的CO、CO_2、NO、NO_2、SO_2、HCl、HBr、HCN 8种典型有毒有害气体进行定量分析。经过合理选择光谱区间、数据预处理、样本筛选以及确定模型参数后,建立PLS回归模型,并对模型回归曲线进行多项式修正。模型中各组分实际浓度与预测浓度的拟合回归系数达到0.99,校正集误差均方根S^(RMSEC)低于15×10^(-6)。利用验证集对模型的预测性能进行检验,样本各组分的预测浓度误差小于满量程的±2%,各组分的预测误差均方根S^(RMSEP)不超过20×10^(-6)。

Abstract： The paper is concerned about a measuring method devel- oped by the present author based on FTIR for measuring the concentrations of components in the toxic gases harmful both to human health and the environment quality. The said experimental system was established based on the use of Fourier transform infrared spectrometer, gas compounding equipment, vacuum pump as well as other devices. To make the quantitative analysis kinds of classic toxic gases with successful, we have chosen eight low concentration, including CO, CO2, NO, NO2, SO2, HCl, HBr and HCN. The best calculated spectrum regions of the 8 components were analyzed in hoping to avoid the interference of vapor and carbon dioxide from the outside environment as the result of a series of trials and appropriate choice. The sample of calibration set has been tested and verified by the method of Mahalanobis distance （M） for the model to be soundly es-tablished. For example, we have deleted two samples due to their being less purposeful for our research goals. Then, two data matrixes were adopted to establish our model, one of which is infrared spectrum data matrix and the other of which is the gas concentration data matrix. And, finally, the partial least square （PLS） regression model was established after choice of an appropriate spectrum region, data pre-treating, standards selection and parameters setting. With the result of each component verified by the PLS model and regulated by the multinomial, the fitting regression coefficient between the actual concentrations and predicted concentrations of each component proved to be higher than 0.99, and the Root-mean Standard Error of Calibration （ SRMSEC ） proves lower than 15 × 10^-6 . With the expected performance of the model well verified by the standards in validation set, the error of the concentration prediction proves lower than ± 2% FS and the Root-mean Standard Error of Prediction （SRMSEe） is lower than 20± 10^-6. Thus, it can be seen that the proposed qualitative analysis of such toxic gases based on FTIR is a new valuable method for gas detection, which can be applied for environmental monitoring, chemical plants＇ quality-inspection, fire-hazards detection, anti-terrorist activities and other areas concerned.