Study On Partial Discharge Characteristics And Recognition Of 10kV Cross-linked Polyethylene Cable Under Oscillation Wave Voltage

Cross-linked polyethylene cable is widely applied in power distribution network. Cable insulation has long been affected by electrical, thermal, chemical and other factors, which will trigger partial discharge(PD) during operation. As an effective diagnosis method for partial discharge test, the oscillation wave voltage method has been adopted by power grid operation management department in recent years. At present, the study of oscillation voltage is limited in detection and location of defect. Therefore, the development of the research of partial discharge characteristics and pattern recognition under the guidance of the oscillation wave voltage has great significance for defining the type of cable insulation defect and taking appropriate measures. The main contents of this thesis are shown as follows:① According to the typical defects occurred during operation of 10 k V XLPE cable, four defects has been made in the cable joint: cusp in semi-conducting layer defect, air-gap in semi-conducting layer defect, metal cusp at high potential defect, too long overlap of stress cone defect; partial discharge comparative tests between the oscillation voltage and power frequency voltage were carried out for four defect model. And the PD raw data of four kinds of defects under the oscillation voltage and power frequency voltage is collected.② The discharge phase-capacity-frequency( j -Q -N) three dimensional statistical map of partial discharge signals under the oscillation wave voltage has been got based on the multiple wave accumulation algorithm; the time domain waveform of partial discharge pulse signal and the three dimensional statistical map are compared and analyzed respectively under the two different voltage mentioned; the time-frequency- energy distribution map is obtained by Hilbert-Huang transform(HHT), it can reflect the variation characteristic of partial discharge signals under the multi-cycle damping oscillation wave voltage.③ The bidirectional two-dimensional principal component analysis algorithms is adopted to extract the features from four kinds of defects in time-frequency-energy distribution map, and then linear discrimination analysis algorithm is utilized to extract the optimal discriminant vectors; by using back propagation neural network(BPNN) classifier and least squares support vector machine(LSSVM) classifier, the optimal discriminant vectors and the traditional two-dimensional map statistical characteristic parameters are taken as the input vector of the two kinds of classifier for four kinds of defect recognition respectively. The results show that the least squares support vector machine with optimal discriminant vectors from time-frequency-energy map as input vector has better recognition effect.