Fault diagnosis of grounding system of high voltage cable circuits using graph attention networks
Sheath currents in HV cable systems often exceed standard values, while no faults were found. Existing fault diagnosis methods for the grounding system heavily rely on theoretical models and neglect the shared grounding points of multiple cable circuits, resulting in unsatisfactory practical perform...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-07-01
|
| Series: | International Journal of Electrical Power & Energy Systems |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S0142061525002340 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Sheath currents in HV cable systems often exceed standard values, while no faults were found. Existing fault diagnosis methods for the grounding system heavily rely on theoretical models and neglect the shared grounding points of multiple cable circuits, resulting in unsatisfactory practical performance. This paper establishes theoretical models of the grounding system in various states of cable circuits with shared grounding grids. Whist differences among the magnitudes of the three phase load currents are assumed negligible, the distribution characteristics of the sheath currents are analyzed and three kinds of correlations between the measured sheath currents and the configuration of the sheath grounding systems are proposed and investigated. A fault diagnostic model based on graph attention networks (GATs) is established with the combination of the three relationships as prior knowledge, enabling the diagnosis of the grounding system without reliance on theoretical models or circuit parameters. The model is validated by field data and simulation data, which demonstrate that the four metrics of the fault diagnostic model are above 90%. The incorporation of GATs and prior knowledge enhances the model’s accuracy by 14.41% in comparison with that of the traditional fully connected neural network model.© 2017 Elsevier Inc. All rights reserved. |
|---|---|
| ISSN: | 0142-0615 |