Study On Background Correction And Overlapped Signals Separation For Chromatogram Peaks Of SF₆ Decomposition Products Under Partial Discharge

Partial discharge (PD) in Gas Insulated Switchgear (GIS) is one of the important factors for arising the fault, and therefore the on-line monitoring PD in GIS has been being researched at home and abroad in order to guarantee the safe operation of GIS.Operation experiences show that PD can induce SF₆ to decompose. Different types of defects may lead to different decomposition process, decomposition rate and gaseous decomposed components. Meanwhile, the studies of analysis for SF₆ decomposition products under PD show that it is inevitable to exist the overlapped signals and the background signals interference in the analysis results based on gas phase chromatograph, infra-red spectrum and so on.In this paper, SF₆ decomposition products was preliminary analysed by gas phase chromatograph based on the developed device of SF₆ decomposed under PD and the designed needle-plate and free metal partials insulation defects. According to the chromatogram data features of SF₆ decomposition products, the overlapped signals and background signals interference were optimized with chemometrics. The main achievements are as follows:The SF₆ decomposed test was tried based on developed device of SF₆ decomposed under PD and the designed needle-plate and free metal partials insulation defects, and analysed by gas phase chromatograph; the component types and contents of SF₆ decomposition products were obtained by gas phase chromatograph with different insulation defects and time quantum to present the conclusion that using the component contents of SOF2 and CF4 as one of the discriminant foundations for the needle-plate and free metal partials insulation defects.The preliminary chromatogram data were resolved with Mallat algorithm of wavelet analysis theory. This method that resetting the high frequency signals, reconstructing low frequency and then correcting the background signals in the original signal was to avoid the data skewness with negative digital points after background correction and achieve the baseline correction; according to the study and analysis of simulated data including curve background and real data of the needle-plate and free metal partials insulation defects, the method is effective; the best wavelet function of the background correction is Db5 and the best resolution times of the background correction is 8 for chromatogram data of SF₆ decomposition products, based on the calculation with different wavelet functions and resolution times.According to the excessively high peak values of SF₆ and Air and overlapped peaks resulted from excessively centralized retention time, the overlapped peaks were separated by curve-fitting. In this method, fitting error E was solved by artificial neural network adjustment formula and peak position, value and width were confirmed by function K(t)=e-k(1-n/T) from wide to narrow with the correction of excessively high peak values and abscissa step size. The study and calculation of simulated data including partial and nearly complete overlapped peak and real data show that this method is effective on the analysis for overlapped chromatogram data of SF₆ decomposition products; according to the correction of excessively high peak values and abscissa step size with different proportions, it shows that the method can be normally operated by correcting the peak values of SF₆ and Air to the 7.5 times of the highest peak value of other products, and the peak position was confirmed more accurately and the parameter was chosen more easily with 1/2 times of the original abscissa step size namely sampling every half minute when separating the overlapped peaks.