Stealthy Noise Injection Attacks Against State Estimation in Interconnected Systems

This brief investigates the stealthy attack design problem for interconnected systems, where the aim of the attacker is to maximize the estimation error covariance of the centralized state estimator by injecting elaborately designed Gaussian noises into the compromised subsystem. The proposed attack strategy can achieve complete stealthiness for local detectors and keep a certain level of stealthiness for the centralized one simultaneously, which differs from the existing attack strategies for interconnected systems. Due to the nature of the attack model and interconnected systems, the stealthiness constraint is so complex that it is difficult to directly solve the formulated optimization problem to obtain the detailed attack strategy. To transform the stealthiness constraint, the eigenvalues of matrices are used to characterize the upper and lower bounds of the corresponding determinant, and then the closed-form optimal attack strategy is derived. Finally, a 1D network simulation example is provided to verify the developed results.