Low Voltage Series Arc Modelling Based on Neural Network Considering Harmonics Load Current

Abstract

Series Arc is one of the electrical fault types in a low voltage power system. Series arc occurred when two points of the same conductor connection have different potential values. It usually happens on a cable with broken insulation. Most people rarely notice the phenomenon of series arc because it may not be visible. When it happens continuously, the temperature around the arc location will increase and potentially cause a fire. The value of the series arc fault current has a similar value as the nominal current and causes protection devices such as circuit-breaker and fuse unable to detect it. Low voltage series arc modeling is necessary to enable protection devices in detecting series arc faults. This experiment conducts low voltage arc modeling on non-linear loads containing THDi values and several line impedance values. The modeling input is the arc voltage, arc current, and arc power before the series arc occurred, and the modeling target is the arc resistance. The modeling method is by using the artificial neural network with feed-forward backpropagation. The experiment shows that the higher the THDi values in the system, the higher the series arc fault current. The modeling results show that the modeling can represent the series arc fault resistance with an MSE value less than 0.04.