A Zero Rest Boundary Characteristic-Guided Attention Fine-Tuning Method for High Impedance Fault Detection in Distribution Networks
IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 2520810-2520810
DOI: 10.1109/tim.2026.3744052 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Huaiyuan Wang und 2 weiteren Autor:innen. Die Arbeit nennt 38 Literaturangaben. Gefördert von: Fuzhou Municipal Major Science and Technology Project.
Abstract (Originalsprache)
High impedance fault (HIF) detection from current measurements can be regarded as the identification of a weak arcing phenomenon from measured waveforms. This task is challenging because HIF currents are usually weak and can be easily confused with transient disturbances caused by load switching, capacitor switching, and magnetizing inrush current. Although emerging deep learning models have effectively improved detection accuracy, their learned detection rules may become physically inconsistent when weak zero rest boundary characteristics are not sufficiently captured from measured current waveforms. To address this problem, a zero rest boundary characteristic-guided attention fine-tuning (ZRBCAFT) method based on a Transformer model is proposed for HIF detection. Temporal attention distributions are extracted from self-attention weights to analyze the learned detection rules. A boundary attention-guiding (BAG) loss is introduced to enhance the attention assigned to weak zero rest boundary regions. A plateau attention-constraining (PAC) loss is designed to constrain excessive attention migration toward nonboundary zero rest plateau regions. In this way, unreasonable learned detection rules for weak zero rest samples can be corrected. Case studies are conducted on a PSCAD distribution network simulation system and a physical 10-kV distribution network test system. The detection accuracy is improved from 98.81 % to 99.47 % and from 94.08 % to 95.32 %, respectively. Both false dismissal and false alarm rates are reduced. The effectiveness and applicability of the proposed method under simulated and real HIF conditions are therefore verified.
Förderung
- Fuzhou Municipal Major Science and Technology Project
Zitieren
APA
Wang, H., Lin, Y., & Chen, Q. (2026). A Zero Rest Boundary Characteristic-Guided Attention Fine-Tuning Method for High Impedance Fault Detection in Distribution Networks. IEEE Transactions on Instrumentation and Measurement, 75, 2520810-2520810. https://doi.org/10.1109/tim.2026.3744052
Vancouver
Wang H, Lin Y, Chen Q. A Zero Rest Boundary Characteristic-Guided Attention Fine-Tuning Method for High Impedance Fault Detection in Distribution Networks. IEEE Transactions on Instrumentation and Measurement. 2026;75:2520810-2520810. doi:10.1109/tim.2026.3744052
BibTeX
@article{Wang2026,
author = {Wang, Huaiyuan and Lin, Yujia and Chen, Qifan},
title = {A Zero Rest Boundary Characteristic-Guided Attention Fine-Tuning Method for High Impedance Fault Detection in Distribution Networks},
journal = {IEEE Transactions on Instrumentation and Measurement},
year = {2026},
volume = {75},
pages = {2520810--2520810},
doi = {10.1109/tim.2026.3744052}
}
EUVIMED