Cyber Resilience for DER-Rich Distribution Grids: Ensemble Detection and Explainable Localization of Unobservable False Data Injection Attacks
IEEE Access · 2026 · 14 · 92116-92134
DOI: 10.1109/access.2026.3704895 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Md Moshiur Rahman und 2 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 36 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: University of Central Florida College of Graduate Studies through the Open Access Publishing Fund.
Abstract (Originalsprache)
Modern cyber-physical power systems increasingly rely on data-driven monitoring and control, expanding the attack surface for unobservable false data injection attacks. In distributed energy resources rich distribution feeders, such attacks can evade conventional residual-based detectors and corrupt operational decisions, motivating detection and localization methods that remain reliable, interpretable, and scalable. However, existing learning-based approaches often exhibit unstable trade-offs between false alarms and missed attacks and typically provide limited phase-resolved localization for actionable triage. This paper proposes a scalable two-stage framework that couples stacked ensemble detection with fusion-based explainable localization. The detection stage uses a calibrated cross-family ensemble that adaptively fuses discriminative XGBoost evidence and Transformer-based temporal anomaly evidence to produce operational attack-confidence probabilities. Upon each detection alarm, the localization stage combines TreeSHAP, DeepSHAP, and reconstruction-error severity evidence, and consolidates them via reliability-weighted logarithmic opinion pooling fusion to identify attacked buses and resolve compromised phases and dominant measurement channels. The framework is validated on modified IEEE 13-node and IEEE 123-node feeders under single- and multi-node, multi-phase unobservable attack scenarios. Results demonstrate near-ceiling detection separability with low false-negative and false-positive rates and well calibrated confidence, while fused explanations yield selective node ranking and compact phase/feature attributions under distributed attacks. Per-sample inference latency profiling further confirms real-time feasibility, with sub-millisecond detection and end-to-end detection-localization latency in the tens-of-milliseconds range on both feeders, enabling online alarming and rapid post-alarm localization in distribution grid cybersecurity monitoring pipelines, thereby enhancing cyber resilience.
Förderung
- University of Central Florida College of Graduate Studies through the Open Access Publishing Fund
Zitieren
APA
Rahman, M. M., Rahman, A., & Sun, W. (2026). Cyber Resilience for DER-Rich Distribution Grids: Ensemble Detection and Explainable Localization of Unobservable False Data Injection Attacks. IEEE Access, 14, 92116-92134. https://doi.org/10.1109/access.2026.3704895
Vancouver
Rahman MM, Rahman A, Sun W. Cyber Resilience for DER-Rich Distribution Grids: Ensemble Detection and Explainable Localization of Unobservable False Data Injection Attacks. IEEE Access. 2026;14:92116-92134. doi:10.1109/access.2026.3704895
BibTeX
@article{Rahman2026,
author = {Rahman, Md Moshiur and Rahman, Airin and Sun, Wei},
title = {Cyber Resilience for DER-Rich Distribution Grids: Ensemble Detection and Explainable Localization of Unobservable False Data Injection Attacks},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {92116--92134},
doi = {10.1109/access.2026.3704895}
}
EUVIMED