Malicious data detection in state estimation leveraging system losses & estimation of perturbed parameters

Abstract

It is critical that state estimators used in the power grid output accurate results even in the presence of erroneous measurement data. Traditional bad data detection is designed to perform well against isolated random errors. Interacting bad measurements, such as malicious data injection attacks, may be difficult to detect. In this work, we analyze the sensitivities of specific power system quantities to attacks. We compare real and reactive flow and injection measurements as potential indicators of attack. The use of parameter estimation as a means of detecting attack is also investigated. For this the state vector is augmented with known system parameters, allowing both to be estimated simultaneously. Perturbing the system topology is shown to enhance detectability through parameter estimation.