Stochastic Defense Against Complex Grid Attacks

Abstract

We describe stochastic defense mechanisms designed to detect sophisticated grid attacks involving both physical actions (including load modification) and sensor output alteration. The attacks are undetectable under a full AC power flow model even assuming ubiquitous sensor placement, while hiding large line overloads. We first show that such attacks can be computed in the case of large transmission systems. The defensive techniques apply network control actions that change voltages in a random fashion in particular by introducing (random) low-rank corrections to covariance matrices.Index Terms-Security, cyberphysical power grid attacks.

Mathematical programming formulations for the alternating current optimal power flow problem

“…False data injection attacks (Chaojun et al 2015;Liu and Li 2017) aim at studying the potential threat of a cyber attack consisting in modifying the voltage measurements of the grid, which would trick the network controller into taking wrong decisions. More complicated (but possible) attacks include a physical alteration of the grid, such as disconnecting some lines or disturbing the load and generation of a zone of the grid (Soltan and Zussman 2017;Bienstock and Escobar 2020).…”

“…Since the actions that the network controller subsequently takes depend on the estimation that he or she makes of the grid status, any possible error on this process can be crucial in order to maintain the system stability. A stochastic defense mechanism that randomly perturbs the power generation in order to unmask the effect of sophisticated attacks is proposed in Bienstock and Escobar (2020).…”

Mathematical Programming formulations for the Alternating Current Optimal Power Flow problem

“…False data injection attacks [18,48] aim at studying the potential threat of a cyber attack consisting in modifying the voltage measurements of the grid, which would trick the network controller into taking wrong decisions. More complicated (but possible) attacks include a physical alteration of the grid, such as disconnecting some lines or disturbing the load and generation of a zone of the grid [75,11].…”

“…Since the actions that the network controller subsequently takes depend on the estimation that he or she makes of the grid status, any possible error on this process can be crucial in order to maintain the system stability. A stochastic defense mechanism that randomly perturbs the power generation in order to unmask the effect of sophisticated attacks is proposed in [11].…”

Learning from power system data stream

“…• "Cyber" and "cyber-physical" attacks on power systems are a venue where fast and effective learning of (changing) stochastics may prove useful in identifying attacks. See [27]. The methodology adopted in this manuscript to identify the quiet periods is explained in Section III-B.…”