False Data Injection Attack Design in Multi-sensor Systems Based on KL Divergence Theory

In this paper, a security issue for Cyber-Physical Systems (CPSs) is considered. We analyse a multi-sensor system equipped with a remote state estimation and a set of detectors. From the perspective of a malicious attacker, one intends to modify the innovation sequence by injecting a Gaussian noise and further destroys the system performance. The state estimation error covariance recursion are derived to quantify the effect of an attack. Furthermore, we study the worst-case false data injection (FDI) attack scenario, where the maximal attack probability is limited by the threshold of Kullback-Leibler divergence detector. Finally, a numerical example is shown to demonstrate the effectiveness of the worst-case FDI attack.