Component concentration analysis of mixed gas is an important part for measurement. In this study a regression model with support vector machine using a data set with 5000 samples is developed and applied to predicting unknown component concentration of mixed gas. Through transformation of kernel function, multidimensional and overlapped spectrum data is mapped into high dimension space, so that regression model of mixed gas is carried out in high dimension space of support vector. Some factors such as unitary process, scan interval, range of wavelength, kernel functions and penalty coefficient C that affect model are discussed. Experimental results show that component concentration maximal error is 1.45%. The difficulties of overlapping feature spectrum, identical method of mixed gas analysis, limit number of training sample and dimension of input spectrum are solved and the model brings important theoretical and applied value.
[1]
Yao Tian-shun.
SVM Based Chinese Text Chunking
,
2004
.
[2]
Vladimir Vapnik,et al.
An overview of statistical learning theory
,
1999,
IEEE Trans. Neural Networks.
[3]
Gao Xue.
A New SVM-Based Handwritten Chinese Character Recognition Method
,
2002
.
[4]
Vladimir N. Vapnik,et al.
The Nature of Statistical Learning Theory
,
2000,
Statistics for Engineering and Information Science.
[5]
Kus Hidajat,et al.
Quality Characterisation of Crude Oils by Partial Least Square Calibration of NIR Spectral Profiles
,
2000
.