Optimal induced-norm and set membership state smoothing and filtering for linear systems with bounded disturbances (Q1298213)

A linear discrete system with uncertain disturbances both in the state equation and in the observation equation is considered. The concepts of set membership estimation, induced-norm estimation and suboptimal induced-norm estimation are introduced through the error operator. The error operator maps the noise vector, which incorporates all the disturbances in the equations plus the disturbance of the initial condition, onto the estimation error. The authors consider separately smoothing problems, where the estimation is performed on the entire sequence of observations, and filtering problems, where estimate of the state vector at times \(k=1,2,\ldots\), are based on the observations prior to \(k=1,2,\ldots\). Necessary and sufficient conditions of solvability of a smoothing problem in terms of Euclidean norm are obtained. Moreover, it is shown that the estimate sets and feasible sets are ellipsoids. It is further obtained that in certain situations the estimate sets and feasible sets are concentric ellipsoids with the same orientation. Explicit equations characterizing optimal smoothing are derived. Although it is shown that in \(l_2\) norm the optimal filter problem can be reduced to calculation of the center of a suitable ellipsoid, computationally it remains very difficult to obtain the solution and the solution can be fully characterized only for the case of energy bounded noise.