Study On Characteristic And Feature Extraction Of SF₆Decompositon Under Partial Discharge

The safe operation of GIS apparatus is crucial to the whole power system.Discovering the internal insulating defects in GIS for soon is the principal method toprevent black out. It is one of the most effective methods to discover those insulatingdefects by detecting the partial discharge (PD) in the apparatus. Since SF6decomposesand generates a series of characteristic products under PD, it is reasonable to detect PDinside the device effectively by analyzing these decomposition components and theirvariations of concentrations. This new method to detect PD causes more and moreconcerns at home and abroad.However, How to identify the type of insulating defects and the severity ofdischarge through detecting the SF6decomposition components is still not far from ablank page. Therefore, the SF6decomposition under PD induced by three typicalinsulating defects: metal protrusion, free metal particles and insulator gap, areconducted in this paper, and CO2, CF4, SO2F2, SOF2and SO2are chosen as thecharacteristic components. Meanwhile, the influences of PD quantity and trace-levelH2O, which are the major influence factors, on the characteristics of SF6decompositionwere studied. Based on this, the characteristic quantity and recognition method used todiagnose the PD fault were constructed, main conclusions are:①The correlation between the SF6decomposition components and the type pf PDwere obtained through the SF6decomposition experiments under different insulatingdefects. The results show that the generating laws of decomposition components aredifferent under different insulating defects. The concentrations of components generatedby PD under mental protrusion defect are much higher than the data obtained underinsulator gap defect; the data obtained under free metal particles and insulator gapdefects fluctuate greatly, which reflects the processes of discharge are not stable.②Two characteristic components concentration rates were constructed to diagnosethe PD faults. The results of study suggest that: the constructed ratioc(CO2+CF4)/c(SO2+SOF2+SO2F2) indicates whether the organic insulating material isinvolved in PD; and c(SO2F2)/c(SO2+SOF2) can be regarded as the PD energy indicator.③The constructed characteristic ratios were coded to obtain the combined codes.For metal protrusions, free metal particle and insulator gap defects, the combined codesare00,01and11successively. The two-level coding tree used to identify the PD type of insulating defect is created based on the combined codes. The comprehensiverecognition rate of coding tree can reach to83.33%.④The influence laws of trace-level H2O on SF6decomposition characteristicswere studied. The mathematical model of influence of c(H2O) on the characteristic ratiowas established according to the relationship between the chemical reaction rate and theconcentration of reactant. The validity of the mathematical model was verified in theexperiment.