Optimal Stealthy Attack under KL Divergence and Countermeasure with Randomized Threshold

摘要

In a cyber-physical system, there are potential sources of malicious attacks that can damage the estimation quality in an underlying network control system. The attacker aims to maximize these damages while the estimator attempts to minimize them. In this paper we define an attack's stealth based on the KL divergence and obtain an optimal attack. Furthermore, we suggest one method in which the estimator may limit the damage to the system while imposing on any attack a probability for it to be non-stealthy.