Disturbance decoupled secure state estimation: an orthogonal projection-based method (Q2103719)

In this paper, the authors study the disturbance decoupled secure state estimation problem of a linear system in the presence of sparse sensor attacks and unknown disturbances, where this concept means that the resulting estimate errors do not depend on the value of disturbances. The authors consider the discrete-time linear system \begin{gather*} x(k+1)= Ax(k)+Bd(k), \\ y(k) = C x(k)+D d(k)+a(k), \end{gather*} where \(x(k) \in \mathbb{R}^n\), \(y(k) \in \mathbb{R}^l\) and \(d(k) \in \mathbb{R}^m\) represent the states, the tensor measurements and the unknown disturbances, respectively. \(a(k) \in \mathbb{R}^l\) denotes the sensor attack signal, which is an \(s\)-sparse vector, that is, there are at most \(s\) nonzero elements in \(a(k)\). The concepts of \(RS\)-observability and \(RS\)-detectability are introduced to analyze the relationship among states, sensors measurements, disturbances and sparse attack signals. In terms of orthogonal projections, the authors present necessary and sufficient conditions to determine \(RS\)-observability and \(RS\)-detectability. The authors prove that \(RS\)-observability, (\(RS\)-detectability), is a necessary and sufficient condition for the solvability of the disturbance decoupled SSR (secure state reconstruction), (asymptotic disturbance decoupled SSE (secure state estimation)). An orthogonal projection-based design procedure is provided to reconstruct (asymptotically estimate) the states under the condition of \(RS\)-observability (\(RS\)-detectability). Finally, the effect of numerical errors on the residuals are analyzed. It is shown that the ultimate bounds of the estimate errors are proportional to the numerical errors.