 |
Center for Nonlinear Studies |
Malicious data attacks in the power grid can lead to incorrect state estimation and in turn affect electricity prices and operational stability int he grid. Such data attacks can be conducted by changing meter measurements or by compromising their secure communication to the control center. We consider an adversary that can corrupt measurement readings using information limited to the structure of the grid and location of measurement meters. More importantly, the adversary does not have access to values of network parameters and the current operating conditions in the grid. First, we discuss an 'undetectable' data attack on a set of critical state variables by the adversary. A polynomial time algorithm is provided to construct the optimal attack vector that requires corruption of minimum number of measurements. Next, we present a new 'detectable' attack model where an adversary produces changes in state estimation despite failing bad-data detection checks at the state estimator. The new attack regime’s significance lies in reducing the minimum sizes of successful attacks to half of that of undetectable data attacks. The complexity of constructing an optimal 'detectable' attack is discussed and two polynomial time approximate algorithms for attack vector construction are developed. To prevent such data attacks, greedy protection algorithms are proposed for two cases, one where the adversary is resource limited and the other where the adversary has no resource constraint. Finally, we discuss 'topology' attacks on the power grid where an adversary changes the values of breaker statuses on transmission lines rather than the meter readings to affect state estimation. We provide conditions necessary for the viability of 'topology attacks' that do not require knowledge of meter readings and discuss their construction.
Host: Misha Chertkov