Two-stage identification of false data injection attacks in power systems via semi-supervised deep learning

Fengrui Liu, Keng-Weng Lao, Yida Xu, Yang Li, Haotian Guo, Xiaorui Hu, Yikun Yin,
Two-stage identification of false data injection attacks in power systems via semi-supervised deep learning,
Applied Soft Computing,
Volume 184, Part B,
2025,
113672,
ISSN 1568-4946,
https://doi.org/10.1016/j.asoc.2025.113672.
(https://www.sciencedirect.com/science/article/pii/S1568494625009834)
Abstract: Previous studies find that malicious data manipulations against the state estimation, such as the false data injection attack (FDIA), can evade the detection of a conventional bad data detector equipped with a measurement meter. This calls for advanced FDIA identification approaches urgently. However, existing data-driven efforts are designed under the assumption that attacks are frequent and the amount of compromised data is comparable to benign data, which may not be realistic. Thus, these approaches deliver unsatisfactory performance under highly imbalanced data in the real world. To overcome this issue, we propose a novel two-stage FDIA identification pipeline, which formulates the problem as global detection and fine-grained localization. Following this framework, we leverage deep support vector data description to distinguish attacks from benign measurements in an unsupervised manner and employ a modified one-dimensional ResNet to locate the attacking aims upon detecting an FDIA. Our approach can overcome existing limitations induced by data-driven methods under infrequent FDIAs, leading to effective and robust FDIA identification. Case studies on IEEE standard 14-bus and 118-bus systems demonstrate the effectiveness and superiority of our approach and validate our findings.
Keywords: Cyber security; False data injection attacks; FDIA identification; Smart grid; State estimation