An Efficient Data-Driven False Data Injection Attack in Smart Grids

Abstract

Data-driven false data injection attack is one of the emerging techniques in smart grids, provided that the adversary can monitor the meter readings. The basic idea is constructing attack vectors from the estimated signal subspace, without knowing system measurement matrix. However, its stealthy performance is significantly influenced by the accuracy of the estimated subspace. Furthermore, it is computationally demanding, because full-size singular value decomposition (SVD) is required for model order selection. In this paper, we propose a truncated SVD based computationally efficient attacking scheme using only the first dominant eigenvector. Both experiment and simulation results are provided to evaluate the performance of the proposed scheme. Compared with the standard false data injection techniques with known measurement matrix, similar stealthy performance is achieved with a reasonable computational complexity.