Data-Driven Stealthy Injection Attacks on Smart Grid

Smart grid cyber-security has come to the forefront of national security priorities due to emergence of new cyber threats such as the False Data Injection (FDI) attack. Using FDI, an attacker can intelligently modify smart grid measurement data to produce wrong system states which can directly affect the safe operation of the physical grid. The goal of this thesis is to investigate key research problems leading to the discovery of significant vulnerabilities and their impact on smart grid operation. The first problem investigates how a stealthy FDI attack can be constructed without the knowledge of system parameters, e.g., line reactance, bus and line connectivity. We show how an attacker can successfully carry out an FDI attack by analysing subspace information of the measurement data without requiring the system topological knowledge. In addition, we make a critical observation that existing subspace based attacks would fail in the presence of gross errors and missing values in the observed data. Next, we show how an attacker can circumvent this problem by using a sparse matrix separation technique. Extensive evaluation on several benchmark systems demonstrates the effectiveness of this approach. The second problem addresses the scenario when an attacker may eavesdrop but only has access to a limited number of measurement devices to inject false data. We show how an attack can be constructed by first estimating the hidden system topology from measurement data only and then use it to identify a set of critical sensors for data injection. Extensive experiments using graph-theoretic and eigenvalue analyses demonstrate that the estimated power grid structure is very close to the original grid topology, and a stealthy FDI attack can be carried out using only a small fraction of all available sensors. The third problem investigates a new type of stealthy Load Redistribution (LR) attack using FDI which can deliberately cause changes in the Locational Marginal Price (LMP) of smart grid nodes. To construct the LR-FDI attack, the Shift factor is estimated from measurement and LMP data. Finally, the impact of the attacks on the state estimation and the nodal energy prices is thoroughly investigated.