|Stochastic Systems Group|
Malicious Data Attacks on Power System State Estimation
Suppose that a malicious adversary enters a power network, seizes several meters, and alters the measurements that are reported to the control center. How can this affect the knowledge at the control center of the system state? If the adversary controls enough meters, it is able to execute an "unobservable" attack, which substantially damages the control center's state estimate. This attack divides the problem into two regimes: the strong attack regime, in which the adversary can execute the unobservable attack, and the weak attack regime, in which it cannot. We present a graph-theoretic characterization of unobservable attacks, allowing efficient calculation of the smallest set of meters necessary to execute the attack. In the weak attack regime, we present a decision theoretic framework that allows us to examine the problem from the perspective of both the control center and the adversary. For the control center, we propose a generalized likelihood ratio detector that outperforms classical bad data detectors. For the adversary, we examine the tradeoff between damaging the state estimate and minimizing the probability of detection.
Problems with this site should be emailed to email@example.com