Identification of state-dependent parameter models with support vector regression

Abstract

A support vector regression approach is presented for the identification of state-dependent parameter ARX models, whose parameters are described as functions of past inputs and outputs. The problem of identifying the state-dependent parameters reduces to a standard support vector regression problem with a kernel function which is defined in terms of the kernels used to represent the individual parameters. Numerical examples show that the support vector method gives accurate parameter estimates for systems which have a state-dependent parameter representation.