Distributed robust attack detection and reconstruction for a class of uncertain nonlinear interconnected CPSs

Distributed Estimation Techniques for Cyber-Physical Systems: A Systematic Review

“…H∞ filter, which has been applied for the detection of biasing attacks on distributed estimation networks [45] and for the joint attack detection and secure state estimation [52]. Luenberger observer, used in [40] for CPSs affected by adversarial attacks on the sensed and communicated information, in [42] for detecting and isolating multiple sensor faults, in [44] for the simultaneous estimation of the state and attack, in [53] with a secure pre-selector, in [47] for the state estimation in networks subject to adversarial attacks.Kalman filter, which has been used in various fields of application, such as fault detection and isolation for systems of systems [37], security of the state estimation in power systems [39], for attack detection in [54], multi-robot tracking [41], monitoring industrial CPSs [43] or estimation of the biofilm growing process in a biological system [50]. …”

“…Luenberger observer, used in [40] for CPSs affected by adversarial attacks on the sensed and communicated information, in [42] for detecting and isolating multiple sensor faults, in [44] for the simultaneous estimation of the state and attack, in [53] with a secure pre-selector, in [47] for the state estimation in networks subject to adversarial attacks.…”

“…This is the case of [45,52], where a unique linear matrix inequality must be solved to find the observer gains. In [44] the authors require to solve decentralised Lyapunov matrix equations to ensure that both the state and the attack is estimated. However, those equations require global information that, in general, is not available in every location, such as the output matrices and Luenberger observer gains of every estimator.…”

“…In fact, there exists no study that has solved this problem using a pure decentralised approach. Other options available are the papers [44,51]. The former, based on a Luenberger observer, is adequate when the system is described as a set of, possibly nonlinear, subsystems.…”

“…To deepen the discussion concerning the required communication (see Table 10), it is noted that in those estimators based on the Luenberger observer, i.e. [40,42,44,47,53], the information exchanges between agents, take place at the same rate as the estimation algorithm. Moreover, the agents exchange the estimated state vector or a sub component of it.…”