Security for control systems: the observability of linear systems under adversarial attacks

Control systems operate in the background to govern many of our infrastructure. The power grid, water distribution networks, water sewage system, mobile sensor networks, unmanned aerial vehicles are just a few examples that rely on control systems for operation. Increasingly, these systems are connected via the internet, which unfortunately, exposes them to cyber attacks.

In this talk, I will present the problem of state estimation for linear systems under attacks, for which the adversary may have control over some of the sensors and inject (potentially unbounded) additive noise into some of the measured outputs. To characterise the resilience of the system, we introduce a new notion of observability, termed ‘observability under attacks’ which specifically addresses the question of whether it is possible to uniquely reconstruct the state of the system by observing its inputs and outputs over a period of time with the understanding that some of the available outputs have been corrupted by the opponent. We provide computationally efficient tests for observability under attacks which amount to testing standard observability for an appropriate finite set of systems. I will then present estimation algorithms that permit state reconstruction in spite of the attacks.