Induced \(\mathfrak{L}_ 2\) norm model reduction of polytopic uncertain linear systems (Q674934)

Using the technique of linear matrix inequalities, the author gives sufficient conditions for model reduction of the uncertain linear system \(x'(t)= A(t)x(t)+ B(t)u(t)\), \(y(t)= C(t)x(t)+ D(t)u(t)\), where the system matrices \(A,\dots,D\) are continuous time-functions which evolve in a convex polytope. These conditions lead to a non-convex feasibility problem. The concepts of generalized Hankel singular values and a generalized balanced realization of an uncertain system of the above type are introduced. Two-sided bounds for the approximation error are derived in terms of the generalized Hankel singular values of the uncertain system. A model reduction algorithm is described, which is based on the alternating projection method for finding a common point in several closed convex sets. An example is considered which shows that the method proposed is superior to some of the existing methods for computing suboptimal values of the approximation error.