On-Line Identification, Model Structure Reduction, and Control Using NARX Models

The Modified Gram-Schmidt orthogonalization algorithm for NARX (Non-linear AutoRegressive with eXogenous input) models is used for the on-line identification and model structure reduction of multi-input multi-output (MIMO) non-linear systems. In addition a novel non-linear adaptive controller, based on the identified model structure is proposed and verified by simulation experiments on a highly non-linear bioreactor model.