Detection against randomly occurring complex attacks on distributed state estimation

With the development of digitization and intelligence of information technology, cyber attacks tend to be much more complicated and intelligent, which can disrupt the normal system operation if no any protection mechanism is implemented. Motivated by the security problem of industrial control system, we study secure estimation problem in which sensors are exposed to hostile communication environment, where the attacker can randomly launch either DoS attacks or data integrity attacks. We design a distributed estimator equipped with a statistical learning based detector for each sensor over wireless sensor network, and derive an optimal gain for the estimator. Moreover, we investigate the relationship between false rate and the chosen confident level of the detector, we also demonstrate the influence of sliding window of the detector on the estimation performance and show the existence of an optimal scaling parameter corresponding to the best estimation performance. Finally, we prove the effectiveness and feasibility of the proposed estimator by some numerical examples.