An Efficient Privacy-Enhancing Cross-Silo Federated Learning and Applications for False Data Injection Attack Detection in Smart Grid
International Journal of Advanced Research in Science Communication and Technology · 01.04.2026 · 58
DOI: 10.48175/ijarsct-32311 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Advanced Research in Science Communication and Technology, online erschienen am 01.04.2026. Verfasst von T. Sireesha und 3 weiteren Autor:innen. Die Arbeit nennt 15 Literaturangaben.
Abstract (Originalsprache)
The accelerating adoption of federated learning as a distributed machine learning paradigm has opened a new dimension of data privacy in interconnected industrial systems. While federated learning allows participating clients to retain raw data locally and transmit only model parameter updates for global training, it remains vulnerable to inference attacks mounted by an honest-but-curious aggregation server. Countermeasures that exist today either demand a trusted third party for key generation, impose computationally expensive operations such as discrete logarithm evaluation every training round, or fail to accommodate the dynamic participation patterns—dropouts and rejoins—that are unavoidable in large-scale industrial deployments. This paper introduces a general-purpose, efficient, and provably private cross-silo federated learning protocol specifically designed to address these shortcomings. The approach pairs a novel double-layer encryption mechanism with Shamir’s threshold secret sharing, arranging the protocol so that secret sharing is invoked only at the key-establishment phase and upon client re-entry rather than at every round. The result is a scheme that achieves cryptographic privacy guarantees without the ciphertext-expansion cost of homomorphic encryption and without the accuracy penalty associated with differential privacy. The framework is applied concretely to the detection of False Data Injection Attacks (FDIA) in multi-area power transmission grids, using IEEE 14-bus and IEEE 118-bus benchmark datasets. Empirical evaluation demonstrates that the proposed federated model attains 93.4% classification accuracy on the 14-bus case and 91.8% on the 118-bus case after fifty communication rounds, staying within two percentage points of a fully centralised ensemble baseline while providing stronger and more principled privacy assurances
Zitieren
APA
Sireesha, T., Nainy², A., Kiran, V., & Sai, V. H. H. N. (2026). An Efficient Privacy-Enhancing Cross-Silo Federated Learning and Applications for False Data Injection Attack Detection in Smart Grid. International Journal of Advanced Research in Science Communication and Technology, 58. https://doi.org/10.48175/ijarsct-32311
Vancouver
Sireesha T, Nainy² A, Kiran V, Sai VHHN. An Efficient Privacy-Enhancing Cross-Silo Federated Learning and Applications for False Data Injection Attack Detection in Smart Grid. International Journal of Advanced Research in Science Communication and Technology. 2026:58. doi:10.48175/ijarsct-32311
BibTeX
@article{Sireesha2026,
author = {Sireesha, T. and Nainy², A. and Kiran, V. and Sai, V. Hari Hara Nadha},
title = {An Efficient Privacy-Enhancing Cross-Silo Federated Learning and Applications for False Data Injection Attack Detection in Smart Grid},
journal = {International Journal of Advanced Research in Science Communication and Technology},
year = {2026},
pages = {58},
doi = {10.48175/ijarsct-32311}
}
EUVIMED