Earth fault distance computation with artificial neural network trained by neutral voltage transients

A novel application of the neural network approach for transient based earth fault location in 20 kV radial power distribution networks is presented. The items discussed are earth fault transients, signal pre-processing, ANN training and the performance of the proposed distance estimation method. The distribution networks considered are either unearthed or resonant earthed. Neural networks trained by the harmonic content of neutral voltage transients were found to be applicable to fault distance computation in the case of very low fault resistance. The mean error in fault location was about 1 km in the field tests using staged faults, which were recorded in real power systems.