The artificial neural-networks-based relay algorithm for the detection of stochastic high impedance faults

The paper presents a practical artificial neural-network based relay algorithm for electric distribution high impedance fault detection. The scheme recognizes the distortion of the voltage and current waveforms caused by the arcs usually associated with high impedance faults. Fourier analysis yields low-order harmonic vectors of the three phase residual current, voltage, admittance and power which are then fed to a neural network. The network is based on either perceptron or feed forward algorithm. The trained network was verified using other several different contingencies on both the training distribution system as well as other representative distribution systems. A high impedance fault model was developed, where the random nature of the are was simulated using MODELS in ATP. (C) 1998 Elsevier Science B.V. All rights reserved.