A Series Arc Fault Diagnosis Method Based on an Extreme Learning Machine Model

In this study, we address the critical issue of accurately detecting series AC arc faults, which are often challenging to identify due to their small fault currents and can lead to devastating electrical fires. We propose an intelligent diagnosis method based on the extreme learning machine (ELM) model to enhance detection accuracy and real-time monitoring capabilities. Our approach involves collecting high-frequency current signals from 23 types of loads using a self-developed AC series arc fault data acquisition device. We then extract 14 features from both the time and frequency domains as candidates for arc fault diagnosis, employing a random forest to select the most significantly changed features. Finally, we design an ELM classifier for series arc fault diagnosis, achieving an identification accuracy of 99.00% ± 0.26%. Compared to existing series arc fault diagnosis methods, our ELM-based method demonstrates superior recognition performance. This study contributes to the field by providing a more accurate and efficient diagnostic tool for series AC arc faults, with broad implications for electrical safety and fire prevention.