Distributed cyber attack detection and physical fault diagnosis for a class of interconnected large-scale systems

In this paper we focus on the distributed cyber attack detection and physical fault diagnosis problem for a class of interconnected large-scale systems (ILSSs). In the proposed scheme, apart from node measurement, edge measurement is also used to construct distributed Kalman filter to estimate the state of each subsystem. The gain matrices of Kalman filter are determined by minimizing the covariance of estimation error in the attack-free and fault-free case, which reduces the false alarm rate of cyber attack detection and physical fault diagnosis. Based on this filter, a bank of adjacency residual generators is constructed to characterize the influence of cyber attack on the edge measurement, and the Chi-square test is used to detect whether the received edge measurements are attacked. At the same time, a local residual generator is constructed for each subsystem to characterize the influence of physical faults on it, and the residual signal is evaluated by variance and directional residual, so as to make distributed fault detection and isolation of each subsystem. It is worth noting that at each step, each subsystem first performs attack detection on the received edge measurements, and then estimates its own state using the attack-free edge measurements and node measurement, which further improves the accuracy of fault detection and isolation. In addition, a sufficient condition that ensuring the mean square exponential boundedness of the estimation error is given. Finally, the proposed scheme is verified by an illustrative example.