Series and Parallel Arc Fault Detection With Dual Sensors Under Simultaneous Power Quality Disturbances Based on S-Transform and Convolutional Neural Network

Abstract

Arc faults are electrical faults cause fires which have the potential to threaten human safety and major losses. In the electric power network, arc faults occur in two main forms of series and parallel which each exhibiting distinct characteristics. To capture the electrical behavior of both types, dual sensing is employed using a combination of current and voltage sensors. The Arc faults may arise simultaneously with power quality (PQ) disturbances in the power grid, making detection more challenging due to the resulting non-stationary signal. To address this issue, this paper proposes Stockwell transform (ST) for accurate detection of series and parallel arc fault under PQ disturbances. ST is utilized to obtain time-frequency analysis in form of image from the series and parallel arc fault signals. Both simulation and a low-cost experimental setup are developed to collect current and voltage signals data. Convolutional neural network (CNN) with ResNet50 model is proposed for classify of arc fault under PQ disturbances using the time-frequency features obtained from ST as input. Eleven types of normal system and disturbances are selected to evaluate the performance of the proposed methods. The results demonstrate that the proposed methods achieve reliable and accurate results in detection of series and parallel arc fault under simultaneous PQ disturbances.