Research On Detection And Data Analysis Of Dissilved Gas In Transformer Oil Based On The Raman Spectrum

As the pivotal equipment of power transmission and voltage conversion,power transformer plays an extremely important role in the power system network,so its fault monitoring is of great significance.Among them,oil-immersed power transformer is widely used in the power industry because of its good heat dissipation,insulation performance and large transmission capacity.When there are overheating or partial discharge faults in oil-immersed transformer,Seven gases（H₂,CO,CO₂,CH₄,C₂H₂,C₂H₄and C₂H₆）are produced by decomposition in oil.Through the analysis of gas composition and content,Transformers can be monitored to repair defects before faults and cooperate with protective actions to ensure the safe and reliable operation of power system network.Compared with the traditional methods of detecting dissolved gases in oil,Raman spectroscopy has many advantages,such as no pretreatment,good repeatability and a wide range of primary detection bands.It has great advantages in the application of on-line diagnosis of transformer operation status.In this paper,based on Raman spectroscopy,the Raman vibration theory of seven characteristic gases dissolved in oil is analyzed.It focuses on two aspects:spectral data preprocessing based on sparsity and quantitative analysis based on partial least squares method,which improves the utilization rate of Raman data and the accuracy of quantitative analysis,and lays a foundation for reliable monitoring of power transformer operation status and accurate judgment of fault types.Firstly,aiming at seven kinds of fault characteristic gas molecules dissolved in oil,the molecular configuration optimization and modeling were carried out by using the calculation method of Gauss09 software based on the second-order correction（MP2）of multi-order perturbation theory.The relationship between different vibration modes and Raman peaks is analyzed,and the theoretical Raman shifts of gas molecules after frequency correction are obtained,which provides a theoretical basis for the assignment of experimental Raman peaks and quantitative analysis.Secondly,Raman spectroscopy experiment platform was built and detection scheme was designed.Raman spectroscopy detection of CH₄ and C₂H₆ mixed gas samples with different component contents was completed.Baseline correction and noise subtraction of Raman spectral data are performed based on piecewise polynomial fitting combined with S-G（Savitzky-Golay）least squares smoothing and BEADS（Baseline Estimation Add Denoising With Sparsity）algorithm;Compared with the general method,BEADS algorithm avoids the problem of choosing suitable fitting order of polynomial fitting method and setting effective moving window in S-G smoothing,obtains spectral data with flat baseline and less noise signal,and achieves effective pre-processing of original Raman spectral data.Finally,the Raman spectral linearity is analyzed,and a single-peak Gauss model is established for the Raman spectrum after pretreatment.The extraction of three characteristic quantities of the mid-peak area,half-peak width and peak height of the Raman spectrum is completed.Combining with theoretical Raman frequency shift,the attribution of Raman spectra peaks(CH₄:2919.12cm^-1、3021.82 cm^-1、C₂H₆:2902.26cm^-1、2956.85 cm-1)in two kinds of gas experiments was determined.By increasing the number of spectral peaks and characteristic parameters in the regression model,a regression model based on multi-peak area of CH₄ gas was established,which reduced the absolute deviation of average value to 0.0018 and improved the accuracy of quantitative analysis.At last,based on the partial least squares method,the regression model between the content of various gas components and the characteristic parameters of multi-spectral peaks is established.At the same time,the effective quantitative analysis of the two gases is completed,which improves the utilization rate of spectral data.It is of great significance for on-line monitoring and fault diagnosis of dissolved gases in transformer oil.