Arc fault detection method based on the WES-GCSNet model

The fault arc is difficult to identify effectively, which may lead to the damage of the power system equipment, and even cause serious safety accidents such as fire. Firstly, a feature enhancement method based on Wavelet energy spectrum (WES) is proposed to increase the difference of current signal between normal operation and arc failure. Then, the grouped convolution (GConv) is proposed to increase the difference of current signal between normal operation and arc failure. GConv and channel confusion (CS) operations are used to improve the conventional convolutional neural network and propose a lightweight convolutional neural network, named GCSNet. Finally, we combine GCSNet with WES to propose a series arc fault detection method based on WES- GCSNet model. The lowest detection accuracy of 98% verified the effectiveness of the model.