Field studies using a neural-net-based approach for fault diagnosis in distribution networks

Abstract

The paper discusses results of studies performed on a new fault-diagnosis method for distribution systems using acquired field data. The effectiveness of the fault-diagnosis method in distinguishing between faulted conditions and system conditions that appear fault-like is demonstrated, for a field-test system, using data recorded at two utility distribution systems. The new method uses two major components: a signal preprocessor and a novel supervised clustering-based neural network which perform fault detection in the presence of arcing, classification of the fault type and preliminary fault location through the identification of the faulted phase. The work represents the first time that a supervised clustering neural network has been used for distribution fault diagnosis.