Neural Network Based High Impedance Ground Fault Detection

Kordowich G, Conrad T, Bluhm JN, Gaube S, Jäger J (2026)

Publication Type: Conference contribution, Conference Contribution

Publication year: 2026

Series: 2026 61st International Universities Power Engineering Conference (UPEC)

Event location: Cagliari IT

Abstract

In this paper, we aim to contribute towards the deployment of neural network based high impedance ground fault detection systems for resonant grounded grids. While significant progress has been achieved in this domain, further advancement is hindered by the scarcity of real-world fault data and the domain shift between simulation and reality. Therefore, a domain randomization approach is utilized to generate a diverse training dataset comprising 20,000 electromagnetic transient simulations. The model is validated using 208 staged fault tests and tested on three months of continuously recorded operational data from three different substations. The system achieves reliable detection of faults with impedances of approximately up to 100 kOhm while maintaining a low false positive rate. Results demonstrate strong generalization across different grid configurations, validating the usefulness of a simulation based training of neural networks for high impedance ground fault detection.

Authors with CRIS profile

Georg Kordowich Professur für Elektrische Energieversorgung Timon Conrad Professur für Elektrische Energieversorgung Johann Jäger Professur für Elektrische Energieversorgung

How to cite

APA:

Kordowich, G., Conrad, T., Bluhm, J.-N., Gaube, S., & Jäger, J. (2026). Neural Network Based High Impedance Ground Fault Detection. In IEEE (Eds.), Proceedings of the 61st International Universities Power Engineering Conference. Cagliari, IT.

MLA:

Kordowich, Georg, et al. "Neural Network Based High Impedance Ground Fault Detection." Proceedings of the 61st International Universities Power Engineering Conference, Cagliari Ed. IEEE, 2026.

BibTeX: Download