[Method of infrared spectrum analysis of hydrocarbon mixed gas based on multilevel and SVM-subset]

The hydrocarbon mixed gas was characterized by multi-component and varied density. In order to deal with the difficulties that can not be actually solved with mass mixture gas spectrum data samples, 15 kinds of subset patterns were determined on the basis of investigations and studies, which needed 5 500 spectrum data samples for training and testing. On the basis of this, a method of hydrocarbon mixed gas infrared spectrum analysis based on 2-levels and 15 SVM-subsets was proposed in the light of the idea of working pattern recognition --> mixture gas analysis --> the final result output. In order to solve the problem of new subset working pattern, the SVM online categorization algorithm based on spectrum data relational rule was used. The experimental results show that the component concentration maximal deviation is 0.41% and the maximal average deviation is 0.04%. The method can be used in other mixture gas infrared spectrum analyses, and has the theoretic and application value.