Distributed resilient fusion estimation for resource-limited CPSs under hybrid attacks (Q6077022)

The authors deal with the cyber-physical systems described by the dynamic equations \[ \left\{ \begin{array}{ll} x_{k+1} = Ax_k +w_k, \\ y_{k,j} = C_j x_k +v_{k,j},j = 1, 2,\dots,N, \end{array}\right. \] where \(x_k\in\mathbb{R}^n\) denotes the system state, \(y_{k,j} \in\mathbb{R}^{m_j}\) are \(m_j\) sensor measurements, \(A\) and \(C_j\) are system and observation matrices, respectively, \( w_k\) and \(v_{k,j}\) are i.i.d. Gaussian noises with zero mean and covariances \(Q\geq0\), \(R_j > 0\), respectively. It is supposed that the pair \((A,C_j)\) and \((A,Q^{1/2}\) are detectable and controllable. Each sensor of the system measures partial state components and sends measurement to the corresponding sink node, which calculates the local optimal estimate (LOE) of the system and send the estimate to the fusion center (FC). For sink node \(j\), the a priori estimate and the a posteriori estimate of the LOE are given by the standard Kalman filter. When the LOE is sent to a fusion center (FC) over wired or wireless channel, resource-constraints are unavoidable. In distributed network systems, each sink node receives the measurement of sensors. A malicious third party can falsify the information of sink nodes. Motivated by the above observations, the authors mainly concentrate on the problem of distributed fusion estimation of cyber-physical systems with resource-constraints in the presence of hybrid attacks. An event-based protector is designed based on the Gaussian mixture model (GMM) clustering algorithm against stealthy attacks. The protector detects attacks depending on the probability distribution of local estimate. By establishing a unified state model which can compensate the information loss caused by hybrid attacks and communication constraints, a mean square optimal event-based distributed fusion estimator (EDFE) is derived. It is applicable to different attack scenarios instead of a specific attack, which is different from the existing results (see, for example, [\textit{B. Chen} et al., IEEE Trans. Cybern. 48, No. 6, 1862--1876 (2018; \url{doi:10.1109/TCYB.2017.2716115}); IEEE Trans. Syst. Man Cybern. Syst. 49, No. 2, 455--468 (2019; \url{doi:10.1109/TSMC.2017.2697450})]). Stability of the proposed algorithms is analyzed. The effectiveness of the proposed EDFE is illustrated with the help of the IEEE4-bus power grid example.