From Accuracy to Assurance: Toward Credible Power Grid Anomaly Detection
Proceedings of the AAAI Symposium Series · 06.11.2026 · 10 (1) · 472-476
DOI: 10.1609/aaaiss.v10i1.43441 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Symposium Series, online erschienen am 06.11.2026. Verfasst von Elisa Zhang und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Anomaly detection can warn power grid operators of abnormal conditions such as cyberattacks and equipment failures. However, many high-performing Artificial Intelligence (AI) models for anomaly detection are black boxes. Their alerts are difficult to trust because the events that trigger them can be hard to diagnose. Utility companies need models that verifiably meet their trustworthiness expectations. This is difficult to achieve since we lack shared definitions and metrics for AI trustworthiness under real-world grid constraints and consequences. In this paper, we argue that credibility provides an evidence-based bridge between model performance and trustworthy deployment. Using griDNA, an autoencoder-based system for cyber-physical power grid anomaly detection, we present a case study showing how explainable AI (XAI) can provide credibility evidence for model alerts by linking them to grid-relevant features. We then outline limitations and future research directions to advance trustworthy AI in the power grid.
Zitieren
APA
Zhang, E., Ridley, L., & Darling, M. (2026). From Accuracy to Assurance: Toward Credible Power Grid Anomaly Detection. Proceedings of the AAAI Symposium Series, 10(1), 472-476. https://doi.org/10.1609/aaaiss.v10i1.43441
Vancouver
Zhang E, Ridley L, Darling M. From Accuracy to Assurance: Toward Credible Power Grid Anomaly Detection. Proceedings of the AAAI Symposium Series. 2026;10(1):472-476. doi:10.1609/aaaiss.v10i1.43441
BibTeX
@article{Zhang2026,
author = {Zhang, Elisa and Ridley, Lily and Darling, Michael},
title = {From Accuracy to Assurance: Toward Credible Power Grid Anomaly Detection},
journal = {Proceedings of the AAAI Symposium Series},
year = {2026},
volume = {10},
number = {1},
pages = {472--476},
doi = {10.1609/aaaiss.v10i1.43441}
}
EUVIMED