GENERALIZED PREDICTIVE CONTROL USING PARAMETER ESTIMATION FROM NEURAL MODEL

JADLOVSKÁ, A.; KABAKOV, N.

Abstract

The purpose of this paper is to show how a feedforward neural network (MLP) can be used for modeling of non-linear process. In this paper is considered the possibility of on-line estimation of the actual parameters from off-line trained neural model of the non-linear process using the gain matrix. This linearization technique is used in algorithm GPC, which requires linear process model. Neural model is linearized by means of instantaneous linearization in each sample. Applying the principle of instantaneous linearization to GPC design give tremendous advantages over the conventional non-linear predictive control design. Practical simulations by language Matlab/Simulink, Neural, NNSID Toolboxes and PredicLib illustrate that GPC strategy using linearization technique by the gain matrix from neural process model produces good performance for predictive control of non-linear process.

Coresponding author e-mail: Anna[dot]Jadlovska[at]tuke[dot]sk

Session: Model Predictive Control