Research on Arc Fault Detection Based on Conditional Batch Normalization Convolutional Neural Network with Cost-Sensitive Multi-Feature Extraction

An arc fault is a potential hazard in power systems, capable of causing serious safety accidents such as fires. Therefore, the timely detection of arc faults and implementation of circuit-breaking measures are crucial for ensuring safety, preventing fires, and maintaining the stable operation of power systems. Although existing studies have made progress in improving the accuracy of their detection, most methods have not proposed effective solutions that address the cost-sensitive problem of feature selection. Thus, a multi-feature method is proposed by combining time-domain, frequency-domain, energy, and spatial features, which are integrated into a CBN (conditional batch normalization) convolutional neural network for detection. The experimental results show that the proposed method outperforms traditional models in terms of its accuracy and misjudgment rate while maintaining a lower computational cost, demonstrating its superior detection performance. This provides an effective improvement for arc fault detection.