Pattern Recognition Method Of Support Vector Machine For Small Sample PD Signals

Insulation defects are closely interrelated to PD sources so that operating states of high voltage electrical apparatus can be clearly monitored by detection devices,and the insulation failure can be simultaneously diagnosed. Therefore, it is important perform the fault diagnosis in the condition of small fault samples.Reviewed many researchers'working results and based on a great deal of literatures published in journals on electrical apparatus fault diagnosis. Seven discharge models were made in accordance with different defects happened in operation. Partial discharge signals were collected with high frequency data acquisition unit, and wavelet de-noised. Ten feature parameters were extracted from partial discharge signal on the basis of the characteristic extraction methods.After analyzing the effect of kernal function on the identification result, Radial basis function was selected as SVM kernel function. Radial basis function parameter and SVM penalty coefficient were found by grid-search method while designing SVM classifier. Feature parameters which were divided into four combinations were used to train classifier, the recognition result showed that statistical parameters were the optimal choice. For small samples pattern recognition, the recognition result showed that PD signal was not suitable to be normalized. Furthermore, after analyzing feature with correlation and class distance, the recognition result was not improved efficiently, but reduced the feature parameter and saved the run time. In order to compare the classification performance of SVM and BP neural network, characteristics were extracted from de-nosed signal and analyzed, the recognition rate of SVM was better than BP neural network.