Identification of multivariable LPV state space systems by local gradient search

Abstract

We present an identification method for multivariable linear parameter-varying (LPV) state space systems that is based on a local parameterization of the system and a gradient search in the resulting parameter space. Both the output error and prediction error identification problems are discussed. Because the method involves solving a nonlinear optimization problem, it is of paramount importance to have a good initial estimate of the model. We show that a recently developed subspace identification method for LPV systems can be used for determining such an initial model.