Single-phase-to-ground fault detection in distribution networks using learnable spectral filtering and channel covariance attention
AIP Advances · 01.11.2026 · 16 (11)
DOI: 10.1063/5.0349809 ↗Kurz gefasst
Fachartikel in der Zeitschrift AIP Advances, online erschienen am 13.11.2026. Verfasst von Jiajun Chao und 4 weiteren Autor:innen. Die Arbeit nennt 40 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Changzhou Jinling Electric Power Engineering & Industry Co. Ltd..
Abstract (Originalsprache)
Reliable identification of single-phase-to-ground faults in unearthed and resonant-grounded distribution networks is difficult because high-impedance and intermittent arcing faults produce weak, nonstationary transients whose amplitudes may overlap those of normal disturbances. This study proposes an end-to-end multichannel classifier that jointly learns frequency-domain representations and channel dependencies from synchronized zero-sequence voltage and current waveforms. A learnable complex-valued spectral filter adjusts discriminative frequency components without imposing a predefined hard-threshold mask. In parallel, channel covariance attention learns projected relationships between voltage and feeder-current features over the complete observation window. Instance normalization reduces sensitivity to physical units and operating-amplitude variations, and the two branches are fused for multiclass fault identification. Experiments on a simulated 10 kV, four-feeder distribution system with 24 joint fault classes yield an accuracy of 98.56%, a macro-F1 score of 0.953, and a Cohen’s kappa of 0.954 under fivefold cross-validation. Ablation and additive-noise tests support the complementary contributions of the spectral branch, channel attention, and normalization. The simulation model reflects selected characteristics of practical distribution scenarios, and the method has also been applied in practical distribution-network environments. The quantitative evidence presented here remains simulation-based; systematic field-recording evaluation and larger-network validation are needed to establish broader engineering generalization.
Förderung
- Changzhou Jinling Electric Power Engineering & Industry Co. Ltd.
Zitieren
APA
Chao, J., Lu, R., Yuan, J., Song, Z., & Lv, Z. (2026). Single-phase-to-ground fault detection in distribution networks using learnable spectral filtering and channel covariance attention. AIP Advances, 16(11). https://doi.org/10.1063/5.0349809
Vancouver
Chao J, Lu R, Yuan J, Song Z, Lv Z. Single-phase-to-ground fault detection in distribution networks using learnable spectral filtering and channel covariance attention. AIP Advances. 2026;16(11). doi:10.1063/5.0349809
BibTeX
@article{Chao2026,
author = {Chao, Jiajun and Lu, Rui and Yuan, Jiawei and Song, Zhengrong and Lv, Zeyuan},
title = {Single-phase-to-ground fault detection in distribution networks using learnable spectral filtering and channel covariance attention},
journal = {AIP Advances},
year = {2026},
volume = {16},
number = {11},
doi = {10.1063/5.0349809}
}
EUVIMED