Novel Attacks against Contingency Analysis in Power Grids
This addresses security risks in power grid operations for grid operators, but is incremental as it builds on known attack vectors in State Estimation.
The paper tackles the vulnerability of Contingency Analysis in power grids to attacks that corrupt State Estimation, showing that adversaries can alter measurements to cause overloading during contingencies without detection, as demonstrated on an IEEE 14 bus system.
Contingency Analysis (CA) is a core component of the Energy Management System (EMS) in the power grid. The goal of CA is to operate the power system in a secure manner by analyzing the system subject to a contingency (e.g., the outage of a transmission line or a power generator) to determine the setpoints that will allow system operation without violation of constraints. The analysis in CA is conducted based on the output from State Estimation (SE), another core EMS module. However, it is also shown that an adversary can alter certain power measurements to corrupt the system states estimated by SE without being detected. Such a corrupted estimation can severely skew the results of the contingency analysis as it will provide a fake model to deal with. In this research, we formally model necessary interdependency relationships and systematically analyze these novel attacks on the contingency analysis. In particular, this research focuses on Security Constrained Optimal Power Flow (SCOPF) that finds out the optimal economic dispatches considering a single line failure (based on the $n - 1$ contingency analysis) and transmission line capacities. The proposed model is implemented and solved to find out potential threat vectors (i.e., a set of measurements to be altered) that can evade CA so that the system will face overloading situation on one or more transmission lines when some specific contingencies happen. We demonstrate our formal model on an IEEE 14 bus system-based case study and verify the results with a standard PowerWorld model. We further evaluate the model with respect to various attacks and grid characteristics.