Feature extraction methods for neural network-based transmission line fault discrimination

The suitability of conventional distance relays to operate correctly under variations in such factors as source impedance, prefault load and fault resistance is still a problem. This paper describes an alternative approach to nonunit protection of transmission lines using artificial neural networks (ANNs). Particular emphasis is placed on describing a methodology whereby the extraction of the input features (from the measured voltage and current signals) to the ANNs is near optimal; with this approach, the results presented clearly demonstrate that the protection technique gives satisfactory performance under a wide variation in practically encountered system operating and fault conditions