Subspace-based state-space system identification

Abstract

The subspace approach to state-space modeling offers numerically reliable algorithms for computing state-space descriptions directly from data. The methods are competitive with respect to traditional prediction-error or instrumental variable techniques, in particular for the high-order multi-input multi-output case. The computations are based on QR-factorization and singular-value decomposition, for which numerically reliable algorithms are available. No numerical search is necessary, nor is a potentially ill-conditioned canonical system description used. The goal herein is to present a state-of-the art algorithm for subspace-based system identification. Some computational details are given, and the user's choices are briefly discussed.