探索了以声光可调滤波器(acousto-optic tunable filter,AOTF)为分光器件的新一代近红外(NIR)光谱仪用于气体检测的可行性,并提出一种多组分混合气体近红外光谱分析的新方法。将一个自制的气室与AOTF-NIR光谱仪配接,从而实现了当前仅限于固体和液体检测的AOTF-NIR光谱仪对气体的检测。实验首先获取并比较了甲烷在不同浓度下的近红外光谱。结果显示,当浓度大于0.1%时,甲烷的吸光度明显地随其浓度的增加而增加。随后参照仪器对甲烷的检测低限设计了甲烷、乙烷和丙烷三组分混合气体样本,并采集了它们的近红外光谱。三种组分气体的定量分析模型由核偏最小二乘(kernel partial least squares,KPLS)回归法建立,模型的预测能力采用检验集的预测均方根误差(root mean square error of prediction,RMSEP)评定。与偏最小二乘(PLS)回归分析效果的对比研究表明,KPLS回归较PLS回归在NIR光谱数据的分析上更具优越性。 The feasibility of a new generation near infrared (NIR) spectrometer with acousto-optic tunable filter (AOTF) as spectroscopic device for gas detection was explored. A multi-component mixed gas near-infrared spectroscopy New method of analysis. A home-made gas cell was interfaced with the AOTF-NIR spectrometer to enable gas detection by AOTF-NIR spectrometers, which are currently limited to solids and liquids. The experiment first obtained and compared the near infrared spectra of methane under different concentrations. The results showed that when the concentration was more than 0.1%, the absorbance of methane obviously increased with the increase of its concentration. Then with reference to the lower limit of detection of methane designed methane, ethane and propane three-component mixed gas samples and collected their near-infrared spectroscopy. The quantitative analysis models of the three component gases were established by kernel partial least squares (KPLS) regression. The predictive power of the model was determined by root mean square error of prediction (RMSEP) assessment. Compared with partial least squares (PLS) regression analysis, it is concluded that KPLS regression is superior to PLS regression in NIR spectral data analysis.