Raman Spectral Signal Processing of Dissolved Gases in Transformer Oil Based on Compound Sparse Derivative Modeling

In order to achieve the accurate quantitative analysis of Raman spectrum data of dissolved gases in transformer oil. A Raman spectral preprocessing method based on compound sparse derivative modeling was adopted for Raman experimental data of CH
₄
and C
₂
H
₆
mixed gas samples. In this method, the Raman spectrum signal is decomposed into sparse peaks, low-pass baseline and stationary white Gaussian noise, the asymmetric penalty function is used to solve the convex optimization problem, and the iterative algorithm is introduced to calculate the parameters. Compared with the traditional method, this method can complete the baseline correction and noise reduction in one step. In order to verify the effectiveness of the method, after preprocessing the original data with this method and traditional method, a partial least square regression model was established between the obtained spectral data and the content of each gas sample, and the average absolute error was introduced to evaluate the effect. The results show that the average absolute errors of the quantitative analysis of the two kinds of gases after compound sparse derivative modeling are methane 0.0034 and ethane 0.0038, which are smaller than the corresponding average absolute errors 0.0084 and 0.0061 after the traditional method. The effectiveness of the compound sparse derivative modeling method in the preprocessing of dissolved gas Raman spectrum data in transformer oil was demonstrated, which lays a foundation for accurate and reliable quantitative analysis of dissolved gas Raman spectrum in transformer oil.