Series Arc Fault Detection Method Based on Load Classification and Convolutional Neural Network

Series arc fault (SAF) is one of the main causes of electric fire hazards. However, the arcing current features are different under different load types, which makes SAF detection challenging. This article proposes a method for detecting SAFs in low-voltage AC distribution networks based on load classification and convolutional neural network (CNN). Firstly, an experimental platform is constructed to simulate arc faults according to the standard IEC 62606. The data is collected from eight different loads, and eight loads are divided into four categories by K-means clustering. Then, a detection method is designed by fusing the CNN and arc fault detection criterion. Finally, an online arc fault detection device (AFDD) is developed by deploying the proposed method to an embedded device, and the accuracy, applicability, and stability of the proposed method are evaluated by the AFDD. The results show that the detection accuracy of the proposed method under trained loads and untrained loads can reach 95% and 96.67%, respectively. Thus, this work can provide a reference for developing AFDD.