Observer-based detection and identification of sensor attacks in networked CPSs

Abstract

This paper presents several observer-based detection and identification strategies to characterize sensor attacks for descriptor type cyber–physical systems (CPSs). In contrast to recent works we do not consider state estimators relying on centralized paradigms, but propose distributed and decentralized state estimators to characterize sensor attacks. We also show a set of sufficient conditions to ensure exponential convergence of the estimation errors in the absence of malicious attacks, along with a bound on the convergence rates. We then design algorithms to solve the attack detection and identification problem for a large-scale CPS. Finally, a series of numerical experiments are performed on the IEEE-6 and 118 bus power systems.