Research On Online Analysis Method Of FTIR Spectrum Of Natural Gas Components

Natural gas is widely used all over the world due to its low-carbon,high efficiency and clean characteristics.With the continuous promotion of the energy industry,the demand for natural gas is increasing.Therefore,the accurate and reliable measurement method of natural gas has attracted the attention of all sectors of society.Fourier transform infrared spectroscopy（FTIR）is used in various fields because of its advantages of fast analysis speed,low cost,and no regular maintenance.However,when using FTIR techniques to detect the composition of natural gas,the spectral absorption peaks of mixed components may overlap as different gas components carry multiple characteristic absorption peaks in the infrared spectrum.Therefore,the difficulty of this technology is in analyzing the collected infrared spectrum.In view of the advantages and disadvantages of spectral data analysis methods that will directly affect the measurement accuracy of gas components,this paper makes an in-depth study of the spectral analysis method of infrared spectra of natural gas components.The main work of this paper is carried out from the following aspects:1.Aiming at the problem that the infrared spectrum of natural gas is easily disturbed by noise,an infrared spectrum preprocessing algorithm is proposed.Firstly,the adaptive iterative algorithm of penalty least squares is used to correct the baseline of the spectrum,and then the spectrum is processed by preprocessing algorithms such as wavelet transform and smoothing.The results show that after baseline correction and using SG（N=19）coupled Sym8 four-layer wavelet denoising algorithm,the mean square error of the map is 2.78e-04,and the preprocessing effect is excellent.2.For the overlapping problem of infrared spectrum characteristic absorption peaks of alkane components in natural gas,an overlapping peak resolution algorithm based on the Gauss linear function is proposed to process the mixed spectrum.First,the mixed spectrum Gauss function model is established by using the principle of mixing spectral additivity.The preprocessed infrared spectrum is used as input,and the measured spectral data and the mean square error of several fitted Gauss peaks are used as the objective function.The Levenberg-Marquardt（L-M）iteration algorithm is used to find the minimum value of the objective function,so that the mixed spectrum can be decomposed into several overlapping Gaussian peaks,making the fitted spectrum approach the original mixed infrared spectrum to the maximum extent.3.To solve the problem that the content of each component of natural gas varies greatly and the components with low content are easy to be missed,the infrared spectrum data is selected from 1800 cm^-1 to 600cm^-1（the high content of CH₄ in this band has little influence on the resolution of the spectrum）band to establish a qualitative recognition model,and the decomposed single component spectrum is classified by the support vector machine（SVM）qualitative recognition algorithm.Based on the relationship between peak area and concentration of characteristic absorption peaks of separated pure component gases,a model for gas component concentration measurement is established by partial least squares（PLS）.The results show that the overlapping peak resolution algorithm can be used to analyze complex overlapping peaks,and can also be applied to infrared spectroscopy of other mixed systems.4.In order to verify the reliability of the spectral analysis algorithm proposed in this paper,an online detection system for natural gas components was set up independently.The infrared spectra of four major alkanes components were taken as the research object,and key technologies such as spectral preprocessing,overlapping peak separation,spectral data type identification,and concentration measurement were used for data analysis.The results show that the method can accurately identify the four major alkanes in natural gas with an accuracy of 100%.At the same time,the online concentration measurement of four alkanes components was achieved,with a maximum relative error of 3.18%.The above results show that the method has good stability and accuracy.