Series Arc Fault Detection in Building Distribution Systems: Physics-Guided Feature Extraction and Entropy-Enhanced Weighted Decision Tree

Series arc fault (SAF) detection is crucial for mitigating electrical fires in building distribution systems and ensuring occupant and property safety. However, existing methods can only detect SAFs in limited or idealized scenarios, often suffering from false alarms and missed detections in more complex real-world systems. To address these issues, this article proposes a novel SAF detection method combining physics-guided feature extraction with an entropy-enhanced weighted decision tree. First, an experimental platform simulating SAFs in building distribution systems is constructed, and the effectiveness of the zero-sequence current coupling signal (ZSCCS) as a detection signal is verified through the systematic analysis and comparison. Second, qualitative characteristics of pulses—including positional distribution, amplitude trends, and occurrence frequency—are analyzed, while theoretical correlations between arcing processes and ZSCCS waveform variations are established through RLC-based circuit modeling. Third, four time–frequency domain features are extracted to quantify fault characteristics in ZSCCS, and a detection algorithm is designed based on these features, an entropy-enhanced weighted decision tree, and arc fault decision criteria. Finally, the method’s effectiveness is validated through both offline and online experiments, as well as anti-interference tests. The results show that the proposed method achieves 98.91% detection accuracy (24960 samples) under complex scenarios involving 13 loads, four circuit topologies, and two arc-generating modes, outperforming existing methods. Furthermore, it has been implemented in a self-developed arc fault detection device (AFDD), demonstrating its potential for engineering applications in building distribution systems.