Detection of high-impedance faults using artificial neural networks.

摘要

Till the present, electric utilities are facing the problem of high impedance fault (HIF) detection on electric overhead distribution feeders. These faults often occur when a bare conductor breaks and falls to ground through a high impedance current path. Most HIFs draw little current, which makes them difficult to detect by conventional overcurrent relays. When such faults are not detected, they create a public hazard and threaten the lives of people. The desire to improve public safety has been the primary motivator for the development of HIF detectors.;The Electromagnetic Transients Program (EMTP) is used to simulate the distribution feeder and generate the training cases for the ANN, which is developed and trained using the "Neural Network Toolbox for MATLAB RTM". The feeder parameters are selected to represent a typical overhead feeder in the distribution network of the Saudi Electricity Company-Eastern Region Branch (SEC-ERB).;Detection techniques based on Artificial Neural Network (ANN) have shown a good capability of detecting HIFs. In this thesis, a multi-layer feed-forward ANN, is designed and trained with the Scaled Conjugate Gradient (TRAINSCG) algorithm to analyze the current and voltage waveforms at the substation 13.8 kV bus and indicate whether the feeder is faulty or not. In addition, the ANN can locate the faulty section of the feeder, identity the faulty phase, and most importantly differentiate between faults and fault-like events, such as normal load switching, with a high degree of accuracy.