Decentralized False-Data Injection Attacks Against State Omniscience: Existence and Security Analysis

Abstract

This paper focuses on how false-data injection (FDI) attacks compromise state omniscience, which needs each node in a jointly detectable sensor network to estimate the entire plant state through distributed observers. To reveal the global vulnerability of state omniscience, we investigate decentralized false-data injection (DFDI) attacks that destabilize the estimation error dynamics but eliminate their influences on the residual in each sensor node. Firstly, the sufficiency and necessity for the existence of such attacks are studied from system eigenvalues and attackable sensors. Secondly, the self-generated DFDI attack sequences independent of system real-time data are designed to achieve the attack objective with elaborate parameters. Especially, the DFDI attack sequences are improved to maintain real values even if the system matrix only has unstable imaginary eigenvalues. Finally, we analyze the secure range for observer interaction weights and the sensor protection scheme to guarantee the security of state omniscience under DFDI attacks. The theoretical results for DFDI attacks are demonstrated with the linearized discrete-time model of an aircraft system.