Hybrid technique for fault location of a distribution line

This paper presents a hybrid technique for fault location in a 11 KV, 30 km distribution line with the R-L load placed at the receiving end. The method proposed in this paper analyzes with post-fault sending end one cycle current signal of the distribution system. Preprocessing of the raw signal is done by wavelet packet transform to acquire the information of frequency sub-bands. Here, four level decomposition is performed by wavelet packet transform having sampled frequency of 30 kHz. Thereafter energy feature is collected from the decomposed coefficient for further preprocessing. From a total set of 16 features, 6 optimal features are selected by a feature selection method during the training process. Train and test matrix are produced by applying various simulation conditions like the fault inception angle, resistance of faults path, location of the fault and fault type. The operating conditions of train data set are made entirely dissimilar from the test data set in order to make the method robust to parameter variations. SVM (Support vector machine) and RBFNN (radial basis function neural network) is used for fault distance prediction. Thereafter the optimal features with the test data set are fed to the SVM (support vector machine) and RBFNN (radial basis function neural network) for fault distance estimation. It was seen from the results that wavelet packet transform with particle swarm optimization based feature selection provides minimum fault location error less than 0.21% as compared to other schemes discussed by various researchers.