Research on Fault Arc Detection Based on Deep Learning

The Arc fault features in low-voltage distribution network are not prominent, and multi feature set recognition is required. In this paper, a method of fault arc recognition based on deep neural network learning is proposed: based on convolutional neural network deep learning, the characteristics of nonlinear load, i.e. fault arc, are automatically recognized and extracted directly from the current waveform, and then the boundary conditions distinguishing normal load conditions from fault arc are established using artificial neural network, Select and optimize the AlexNet network architecture to build a deep learning based fault arc recognition model; Select the optimal parameter combination to implement the model training strategy. Finally, based on the constructed fault arc test platform with a computer RAM of 16.00GB and NVIDIA GeForce GTX1060 graphics card, the operating data under various working conditions were sampled and analyzed using the above method. The analysis results accurately identified the occurrence of fault arcs, verifying the effectiveness of the proposed method.