Mechanism-guided temporal representation and risk-aware identification method for high-impedance ground faults in distribution networks under multiple types of confusing disturbances

Authors

• Huaizhou Liu, College of Electrical & Information Engineering ， Changsha University of Science & Technology ， Changsha 410114 ， China
• Junqin Liu, College of Electrical & Information Engineering ， Changsha University of Science & Technology ， Changsha 410114 ， China
• Feng Deng, College of Electrical & Information Engineering ， Changsha University of Science & Technology ， Changsha 410114 ， ChinaFollow

Keywords

distribution network, high-impedance ground fault, zero-sequence current, mechanism-guided, multi-scale residual attention temporal network, risk-aware LightGBM

Abstract

To address the difficulty of representing the weak features of high-impedance ground faults （HIFs ） under multiple types of confusing transient disturbances，as well as the insufficient consideration of missed-detection risk，a mechanism-guided identification method based on multi-scale residual attention temporal representation and a risk-aware light gradient boosting machine （LightGBM ） is proposed.First，using the original zero-sequence current as input，a multi-scale residual attention temporal convolutional network is constructed based on the HIF waveform mechanisms of random arcing，local spikes，half-cycle asymmetry，and nonstationary fluctuations to adaptively extract deep discriminative features at different time scales.Second，the deep temporal features are input into LightGBM，and protection-oriented class risk weights are introduced so that the classification boundary is adjusted toward reducing HIF missed detections.Finally，simulation and a 10 kV full-scale test are conducted to verify the effectiveness of the proposed method.The results show that the proposed method achieves an overall identification accuracy of 98.19% and an HIF recall of 99.17% on the simulation validation set.On the independent test set from the full-scale test，the overall identification accuracy and HIF recall are 97.69% and 98.00%，respectively. The proposed method effectively reduces the risk of HIF missed detection under multiple types of confusing disturbances.

DOI

10.19781/j.issn.1673-9140.2026.04.001

First Page

1

Last Page

11

Recommended Citation

Liu, Huaizhou; Liu, Junqin; and Deng, Feng (2026) "Mechanism-guided temporal representation and risk-aware identification method for high-impedance ground faults in distribution networks under multiple types of confusing disturbances," Journal of Electric Power Science and Technology: Vol. 41: Iss. 4, Article 1.
DOI: 10.19781/j.issn.1673-9140.2026.04.001
Available at: https://jepst.researchcommons.org/journal/vol41/iss4/1