Feature Extraction based Technique for Fault Classification in Power Distribution System

In this paper, a hybrid feature extraction and classification technique is proposed for fault detection in a power distribution network. The scheme uses the wavelet packet transform (WPT) technique for signal analysis and the support vector machine (SVM) technique is used for fault classification. The WPT is used to extract fault current statistical feature from 1/4 of the post fault current signal. The extracted features are subsequently utilized to train and test the SVM scheme for fault classification. Prior to the classification of the fault, the genetic algorithm (GA) technique is used to determine the optimal parameters of the SVM classifier. The results show that the fault classification on distribution line can be determined with high accuracy even for high impedance faults. The classification results are improved from 93.8 % to 98.8 %, when using the WPT technique. The proposed method is tested on a machine learning platform called ORANGE.