High impedance arcing fault detection on underground distribution cables using artificial neural network

Abstract

High impedance arcing fault cause by insulation failure is a nuisance because it is undetectable by conventional overcurrent relay. Because of the similarities of the high impedance arcing fault and other fault- like transient event, an algorithm that can discriminate between them will have to be obtained. In order to obtain transient waveforms of the fault events, digital simulations are performed using PSCAD/EMTDC for different variations of faults types and fault location. A local TNB commercial distribution feeder with various types of loads is use for simulation purposes. Training data set and testing data set are obtained by extensive simulations of the distribution system using an arc model. The backpropagation training algorithm is use to learn the rules for the solution of the problem by observing the sets of input-output pairs. The results indicated that the neural network is able to reach the solution of the problem. The detection algorithm is able to identify high impedance arcing faults with a high percentage of success rates.