Bad data detection in smart grid for AC model

Failures in power grids have proven to be catastrophic. Continuous system monitoring is essential for security and reliable operation of power grids. State-space models are used to estimate the states of the system from available measurements. A malicious user can introduce an error in measurements or bad data can be introduced at different points in the grid resulting in unpredictable behavior of the control algorithms in SCADA systems, which use the state information to make decisions. Bad data detection is part of the state estimation process, but if the attacker has complete knowledge of the system and access to large enough number of measurements, these attacks can be made undetectable. Various techniques to introduce such undetectable attacks have been discussed in literature with focus on DC models. A data history based heuristic algorithm was proposed recently that can detect such attack. However, this technique fails when data attack model is a slow ramp. We propose a novel technique based on rate of change of largest singular value of the data matrix that can detect even slow attacks on the system with focus on AC models.