Detection And Data Analysis On Dissolved Gas In Transformer Oil Based On The Raman Spectrum

Power transformers are very important in power system.To get the accurate running state of the power transformer and carry on effective maintenance in time,this paper uses the Raman spectroscopy detection method which belongs to the DGA（Dissolved Gas Analysis）to detect the dissolved gases in transformer oil.The Raman traits of the gases were obtained through the Raman simulation and with the help of the Gaussian 09 W software.After the analysis on the linetypes of the Raman spectrum,more accurate spectral data processing methods were used,and a new characteristic parameter was found useful.The precision of the Raman quantitative analysis was improved under the existing experimental conditions,which has laid a foundation for the reliable assessment of the power transformers.Firstly,the molecular configuration optimization and the simulations of the 7 kinds of fault gases（CH₄、C2H6、C2H₄、C₂H₂、CO、CO2、H2）have been done based on the B3 LYP method of the density functional theory（DFT）.Combined with the Raman experimental platform,Raman detection of the seven kinds of fault gases was made.Secondly,after the baseline deduction,smoothing and denoising,the two kinds of well-fitting models（Voigt fitting and GaussAmp fitting）were respectively used on the Raman spectral lines of the CH₄ and the C₂H₂.Comparisons between the Voigt fitting and GaussAmp fitting were done.Under the condition of the experiment in this paper,the difference value between the two kinds of fitting area of the spectral peak is nearly250 a.u·cm-1.Both the AIC values and the adjusted determination coefficients can prove that the GaussAmp fitting is able to do a better job in the Raman quantitative analysis of the dissolved gases in transformers.Then this paper looked for new characteristic parameters from the view of the descriptive statistics aiming at improving the precision of the Raman spectroscopy quantitative detection on dissolved gas in transformer oil under the existing experimental conditions.The descriptive statistics parameter kurtosis is found helpful when used as a new parameter in Raman spectrum analysis.Quantitative analyses of the acetylene gas have been made respectively based on the GaussAmp fitting and the kurtosis.After comparison the quantitative analysis based on kurtosis has been found rather better and without pretreatment of the experimental data.Finally,Based on the nonparametric regression model,the statistical parameter（kurtosis）was processed together with the original parameter（peak area）to improve the accuracy of the Raman quantitative analysis on CH₄ in transformer oil,and preferable results were obtained,which provides reliable technical support for the on-line monitoring of the power transformers.