Identification of Approximative Nonlinear State-Space Models by Subspace

Authors:	Schrempf Andreas, University of Applied Sciences - Upper Austria, Austria Verdult Vincent, Delft University of Technology, Netherlands
Topic:	1.1 Modelling, Identification & Signal Processing
Session:	Subspace Methods
Keywords:	Subspace identification, state-affine systems

Abstract

A subspace identification algorithm for state-affine state-spacesystems which allows to approximate nonlinear systemsarbitrarily well is derived. The proposed algorithm depends on anapproximation step where a detailed approximation error analysis isprovided. A special case is presented in which this approximation errorvanishes. To tackle higher-order systems and ill-posed problems aregularized kernel method is proposed. The algorithm is evaluatedby a simulation study.