Open AccessResearch and application of SVM analysis model for spectral fingerprint of mixed gas
Author(s) -
Zh MuRong,
Jun Guo,
Y. X. Zhang,
Yu-Gang Jiang,
Zh G. Yu,
Peng Bai
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1721/1/012054
Subject(s) - fingerprint (computing) , support vector machine , infrared , sample (material) , sensitivity (control systems) , principal component analysis , computer science , analytical chemistry (journal) , chemistry , artificial intelligence , optics , physics , engineering , electronic engineering , chromatography
The infrared spectrum analysis method has the characteristics of non-contact, high-speed and no consumption of sample, There is some useless interference information that is not sensitive to the analysis results in the infrared spectrum data sample. Combining support vector machine with infrared spectrum analysis, the band of spectrum data that is sensitive to gas concentration is selected by experiment. The useless, insensitive and disturbing information is removed, and the useful information is retained. And then the infrared spectrum analysis model based on support vector machine is used to complete the output of gas concentration of mixed gas components. Based on Lambert’s theorem and using the known spectrum data samples as input in experiment, the sensitivity analysis of the mixed gas spectrum data samples is carried out by using the variable sliding window technology which is similar to the center wavelength and band-pass filter, and the start point and band width of the band are both variable. The band with small error is selected as the spectrum data fingerprint series to edit and splice, and then the mixed gas data sample with the fingerprint characteristics of the mixed gas component gas spectrum data is formed. On this basis, a support vector machine based infrared spectrum analysis model of mixed gases is established. The analytical model includes two processes: training and verification. The preconfigured mixed gas samples that are known concentration is fingerprint edited and processed after scanning by the spectrometer. Then the gas samples are formed the spectral data samples which can be considered as the input samples of the analytical model. The model is trained to determine the support vector and corresponding weight of the analytical model based on the spectral data samples; The fingerprint characteristic data samples of the unknown-concentration mixed gas spectrum are tested by the analytical model that has determined parameters. Then, the gas concentration of the mixed gas components is obtained. The aspects of spectral data’s fingerprint feature extraction and the selection of the model’s main parameters are researched, through experiment, the methods and processes of fingerprint feature editing and splicing extraction of spectral data are studied, and the influence of the model’s parameters on the analysis results is optimized. The experimental results show that under the condition of optimizing the model’s parameters, selecting the fingerprint features of the spectral data reasonably and constructing a new spectral data sample with the fingerprint features of the mixed gas component can help eliminate the measurement cross sensitivity caused by the cross overlap of the absorption spectra of the mixed gas component, and improve the calculation efficiency and the analysis accuracy, which has theoretical and practical value.