(A Little) Ignorance is Bliss: The Effect of Imperfect Model Information on Stealthy Attacks in Power Grids

Stealthy attacks on control systems have the potential to cause serious harm, as they are not readily detectable by any intrusion detection system. However, it is often the case that neither the controller nor the attacker can gather perfect information about the system model. We show that while a small mismatch between model and reality can easily be managed by a robust controller, an attacker's imperfect knowledge of the system can thwart the stealth of the attack. This opens the door to a whole new class of defense mechanisms, which focus on maximizing the attacker's uncertainty about the system while maintaining the controller's uncertainty within the bounds of its robustness. We demonstrate our findings on the simple quadruple tank control process and on a realistic power grid model, showing that the time to detect a stealthy attack depends on the attacker's level of uncertainty about the model.