Arc Fault Detection and Classification in DC Microgrid Using Deep Neural Network

Abstract

High resistance DC arc faults in the DC microgrid can create serious damage to the microgrid and put the operator's safety in danger. If it is not quickly found and eliminated. Available pattern-based fault identification approaches do not perform as expected due to the nonperiodic nature of arc fault and the presence of multiple switching converters. This research suggests using Wavelet Transform (WT) in conjunction with deep neural networks to detect arc faults in DC microgrids. Multi-Layer Perceptron (MLP)/Dense neural networks and Convolution Neural Networks (CNN) have been employed in this proposed methodology. The MATLAB simulation using the Cassie arc model is developed, and simulation results. According to the simulation results, MLP and CNN have respective arc fault detection accuracies of 95.5% and 96.4%. The result also shows that CNN performs better in various degrees of noisy signal conditions.