High impedance fault detection in power distribution systems using wavelet transform and evolving neural network

This paper concerns how to apply an incremental learning algorithm based on data streams to detect high impedance faults in power distribution systems. A feature extraction method, based on a discrete wavelet transform that is combined with an evolving neural network, is used to recognize spatial–temporal patterns of electrical current data. Different wavelet families, such as Haar, Symlet, Daubechie, Coiflet and Biorthogonal, and different decomposition levels, were investigated in order to provide the most discriminative features for fault detection. The use of an evolving neural network was shown to be a quite appropriate approach to fault detection since high impedance faults is a time-varying problem. The performance of the proposed evolving system for detecting and classifying faults was compared with those of well-established computational intelligence methods: multilayer perceptron neural network, probabilistic neural network, and support vector machine. The results showed that the proposed system is efficient and robust to changes. A classification performance in the order of 99% is exhibited by all classifiers in situations where the fault patterns do not significantly change during tests. However, a performance drop of about 13–24% is exhibited by non-evolving classifiers when fault patterns suffer from gradual or abrupt change in their behavior. The evolving system is capable, after incremental learning, of maintaining its detection and classification performance even in such situations.