Research On Feature Extraction And Qualitative Identification Method For The Fourier Transform Infrared Spectra Of Atmospheric Gaseous Pollutants

Accurate and rapid measurement of the types and concentration of gaseous pollutants is conducive to the study of the formation mechanism of air pollution and the formulation of pollution prevention policies and regulations.Fourier Transform Infrared Spectroscopy(FTIR)is an efficient environmental air monitoring technology,which allows real time online measurement of trace gases in the air,the problems of spectral overlap and data multicollinearity in measured gas spectra seriously restrict the accuracy of qualitative identification and quantitative analysis of atmospheric spectra.In order to solve these problems,this dissertation conducts in-depth research on feature extraction of gas infrared spectra and qualitative identification of overlapped spectra,and proposes a series of new gas spectral processing algorithms.The main research work of this dissertation can be summarized as follows:1)A spectral denoising algorithm combing improved threshold lifting wavelet transform with adaptive filter is proposed.In view of the unknown statistical features of the noise in measured atmospheric spectra and the similar characteristics of some absorption peaks and noise in frequency-domain,the improved threshold lifting wavelet decomposition is used to denoise the spectra partially and the decomposed high frequency signal is used as the noise reference input of the adaptive filter to make a second denoising of the spectra.The proposed algorithm makes full use of the fast decomposition of lifting wavelet and the tracking and feedback ability of adaptive filter to unknown noise.It can denoise the spectra quickly and retain more details of absorption peak.Experimental results verify the effectiveness of the algorithm.2)An iPLS-MC characteristic wavelength selection algorithm combining interval partial least squares(iPLS)and Monte Carlo sampling(MC)is proposed.In the light of the influence of redundant information and multicollinearity in spectral data on the prediction accuracy of quantitative correction model,the algorithm uses iPLS method to pre select the characteristic wavebands,considering the interplay of wavelengths and the effect of combined modeling on the prediction accuracy of the model,Monte Carlo sampling is used to select the best characteristic wavelengths combination.Modeling by the characteristic wavelengths selected by this method can effectively improve the prediction accuracy of the model and reduce the complexity of the model.Experimental results verify the effectiveness of the algorithm.3)A fast-qualitative identification method of overlapped spectra based on improved independent component analysis(ICA)is proposed.Aiming at solving the problem that the traditional spectral qualitative identification method cannot effectively identify the overlapped broad linewidth spectra,this method studies the similarity between the mixed gas absorbance spectra and the instantaneous hybrid system model,and applies the ICA method to separate the overlapped spectra.The qualitative identification results of the overlapped spectra are obtained by comparing the separation results with the standard spectral database information.The improved ICA algorithm accelerates the process of spectral separation by using the fifth-order Newton iteration,which is beneficial to the rapid qualitative identification of online measurement systems.The experimental results verify the effectiveness of the method.4)The spectral quantitative analysis method based on automatic baseline correction and nonlinear least squares(NLLS)is investigated.With regard to the influence of the baseline drift on the spectral quantitative analysis,the method uses the fourth-order polynomial to fit the background spectra which can automatically correct the baseline,the NLLS concentration inversion method is used to make quantitative identification of the spectra after baseline correction.The method can effectively suppress the spectral baseline drift while retaining the broad linewidth morphology in the spectra.The NLLS concentration inversion method uses the molecular spectral absorption data combined with the instrument parameters to fit the measured spectrum,it can quickly and accurately obtain the concentration information from the mixed gas spectra.The experimental results verify the effectiveness of the method.