Optimal parameter identification for reduced-order discrete-time modeling of time-varying systems

Abstract

A generalized approach for optimal model reduction of discrete systems is proposed. The performance measure is quadratic with positive-definite weighting. Minimizing the associated Hamiltonian with respect to the reduced-order model (ROM) parameters yields an optimal result. Assuming a bounded uniform realization of the full-order system, it is shown that the corresponding Hessian matrix is always positive definite provided the ROM is also uniform