Kullback-Leibler Divergence Based Stealthy Deception Attacks Against Multi-Sensor Remote State Estimation Under Limited Resources

Abstract

Cyber-physical systems enable the remote monitoring and control of physical plants by seamlessly integrating computation, communication, and control. Unfortunately, security implications posed by deception attacks have become increasingly prominent. Designing deception attacks is fundamental to analyzing system vulnerabilities, yet their impact is constrained by attack stealthiness and limited attack resources. Thus, an $\varepsilon$-stealthy deception attack scheme against multi-sensor remote state estimation is proposed in this paper, where only partial sensor residuals are manipulated under limited attack resources. Kullback-Leibler divergence is used to quantify the $\varepsilon$-stealthiness, and a constraint on the covariance of the compromised residual is established. The estimation error covariance of the compromised system is then derived as the attacked objective. Next, the worst-case covariance of the compromised residual is determined by maximizing the trace of the system estimation error covariance under the residual covariance constraint. Based on this, a constraint optimization problem is formulated to derive a worst-case attack strategy. Finally, simulation results are provided to validate the effectiveness of the proposed attack scheme.