Attack Resilient State Estimation by Sensor Output Coding

Designing control systems with a level of resilience against malicious attack has become an important problem because a number of attack incidents recently occurred resulted in significant damages. Resilient state estimation refers to a method of correctly estimating plant states in spite of attacks on sensors. The underlying principle for the majority of existing resilient state estimation methods is to take advantage of redundancies in sensing. More specifically, in order to tolerate (i.e., correctly estimate the plant states) attacks on up to q sensors, the plant must satisfy 2q-redundant observability. In this work, we propose a sensor output coding based resilient state estimation that tolerates the same level of attacks but requires the plant to satisfy a relaxed condition of q-redundant observability. The coding based mechanism identifies the attacked sensors and invokes a state estimator that uses only intact sensors. This enables a correct state estimation even if more than half the sensors are corrupted, which is not possible in most existing works. To demonstrate the effectiveness of the proposed method, simulation results are presented.