Model reference control of nonlinear TITO systems by dynamic output feedback linearization of neural network based ANARX models

Abstract

A dynamic output feedback linearization technique for model reference control of nonlinear TITO (two-input two-output) systems identified by an Additive Nonlinear Autoregressive eXogenous (ANARX) model is proposed. ANARX structure of the model can be obtained by training a neural network of the specific restricted connectivity structure. Linear discrete-time reference model is given in the form of transfer matrix defining desired zeros and poles of the closed loop system. NN-based ANARX model can be linearized by the proposed linearization algorithm thus that the transfer matrix of the linear closed loop system corresponds to the given reference model. The proposed linearization algorithm can be applied to control of a wide class of nonlinear SISO and TITO systems. The effectiveness of the proposed control technique is demonstrated on numerical examples.