Identifying covert data-manipulators in power system estimation loops

Abstract

We develop a set of algorithms for identifying covert data manipulators in distributed optimization loops for estimating oscillation modes in power systems. The fundamental set-up for the optimization is based on Alternating Direction Multiplier Method (ADMM), implemented via message passing between a set of local estimators and a central coordinator. Some of these local estimators are assumed to be compromised by malicious attackers that send incorrect values of their local estimates to the coordinator. Even a small amount of such bias can easily destabilize the ADMM loop. In our first algorithm, we catch the identity of these attackers by employing the standard ADMM but adjusting the value of the penalty factor used in the update of the primal variable. We show that this adjustment can amplify the attack signature, and help in identification, especially when the attack magnitude is small. In our second algorithm, we employ a Round-Robbin variant of ADMM, and catch the manipulators by simply observing the evolution of the dual variable. We illustrate the results using simulations of the IEEE 68-bus power system model.