A physics-guided dynamic graph attention network for false data injection attacks detection and localization in smart grids

Hu, Xiang, Qi, Qi, Wu, Jianzhong ORCID: https://orcid.org/0000-0001-7928-3602, Zhang, Deying, Yang, Haoran and Qi, Bing 2026. A physics-guided dynamic graph attention network for false data injection attacks detection and localization in smart grids. IEEE Transactions on Smart Grid 10.1109/tsg.2026.3706685

Full text not available from this repository.

Abstract

The digitization and interconnection of modern distribution networks pose a severe threat from false data injection attacks (FDIAs). By bypassing conventional detection mechanisms, FDIAs can misguide dispatch decisions and directly jeopardize operational security. Therefore, accurate detection and source localization of stealth attacks are crucial. However, most existing methods neglect the inherent physical constraints of distribution networks, resulting in insufficient localization accuracy and reliability. To address this gap, this paper proposes a physics-guided dynamic graph attention network (PG-DGAT) for stealth FDIAs detection and localization, which deeply embeds the network physical principles into the learning framework of graph neural networks (GNNs). First, adaptive edge-weights are generated from power-voltage coupling to dynamically characterize abnormal variations in nodal correlations induced by FDIAs in the state estimation. Second, a physics-guided message passing mechanism is designed, where the adaptive edge-weights are incorporated into attention computation and a gated recurrent unit (GRU) is employed for feature updating, effectively suppressing noise interference and enhancing the capability to capture low intensity attacks. Finally, physical constraints of distribution networks are incorporated into the loss function to ensure that the model outputs comply with physical operational principles. Simulation results on IEEE 13-bus and 123-bus systems demonstrate that PG-DGAT outperforms representative benchmarks across multiple metrics including accuracy, precision, and recall, while also exhibiting compelling interpretability and practical application potential.

Item Type:	Article
Date Type:	Published Online
Status:	In Press
Schools:	Schools > Engineering
Publisher:	Institute of Electrical and Electronics Engineers
ISSN:	1949-3053
Last Modified:	06 Jul 2026 15:15
URI:	https://orca.cardiff.ac.uk/id/eprint/187958

Actions (repository staff only)

Edit Item