Abstract
We consider combined data and physical attacks on power grids, motivated by recent events and research. We consider a setting where an attacker may alter the topology of a power grid by removing lines and may also alter the load (demand) of some nodes; simultaneously the attacker interdicts data flowing to the control center. We use the PMU model of data that provides high-fidelity AC power flow data (voltages and currents) The goal of the attacker is to provide data that paints a completely safe picture for the grid which is consistent with the net load change, while at the same time disguising large line overloads, a fundamentally dangerous situation that may lead to a cascading failure. We provide a computational procedure that efficiently computes sparse attacks even on cases of large grids.