Neural network based earth fault detection and location on a fourth rail DC railway

Abstract

This paper describes the application of neural networks in earth fault detection and location on a fourth rail DC railway power supply system. A multi-layer perceptron (MLP) network is used with the Leventberg-Marquardt algorithm as the training algorithm. The neural network based fault detector uses 600 Hz harmonic values of voltages and currents at the DC side of rectifiers as the inputs of the neural network. To get the training and testing data, simulations have been conducted to address different complex fault situations. Results show that the neural network based fault detector is fast and accurate. Further work, including more field tests to build on earlier limited tests, will be carried out to investigate the implementation of the neural network based detector for the fourth rail system in real life.