Research on series arc fault detection based on GAF and improved convolutional neural network

Series fault arcing in low-voltage AC circuits occurs due to minimal current variations and its concealed nature, making it difficult for overload protection systems to detect. This undetected arcing can readily cause damage to electrical equipment and has been identified as one of the primary causes of electrical fires. With the increasing variety of electrical loads, the large number of non-linear loads access makes the traditional fault detection based on the current timing signal poor performance. In order to solve this problem, this paper proposes a series fault arc detection method based on Gram Angle Field (GAF) and improved convolutional neural network. Firstly, based on the constructed low-voltage AC series arc fault experimental platform, the measured point current data under the normal and faulty two states of the line under multi-load conditions are obtained. In order to make full use of the advantages of deep learning in image vision, data noise reduction and data visualisation were achieved by VMD sample entropy noise reduction algorithm and GAF algorithm, respectively. Subsequently, an improved convolutional neural network model was constructed by introducing multi-scale residual blocks. The experimental simulation results show that the recognition accuracy of series fault arc reaches 100%, which verifies the effectiveness and accuracy of the proposed method.