Adaptive Secure State Estimation for Cyber-Physical Systems With Low Memory Cost

Abstract

Computer memory occupation is an important issue in secure estimation of cyber-physical systems. The existing secure state observers, which estimate system state from sparsely corrupted measurements, often need to run combinatorial subestimators such that they demand tremendous memory occupation. To this end, this article proposes two adaptive observer structures with the low memory cost. First, a purely adaptive structure is developed under the constraint called $P$ -problem, which allows the observer to automatically search the attack model and update the observer gain matrices. To remove the $P$ -problem constraint, an event-triggered algorithm is further developed by trading off the estimation performance. These two strategies substantially reduce the memory occupation by online updating observer gain matrices rather than offline calculations. Another advantage of these schemes is the use of saturated output injection to mitigate the transient performance degradation in the search phase of the attack model.