Intelligent Modeling and Predictive Control of Non-linear System Based on Forward Neural Model

Abstract

The paper provides two approaches for design of Generalized Predictive Control (GPC) algorithm for nonlinear and time-variant dynamic system. In classical approach of GPC strategy is considered method of instantaneous linearization for calculating of linearized model parameters from known analytic description of nonlinear system. The other purpose of this paper is to show an intelligent approach in which a feed-forward neural network (Multi Layer Perceptron-MLP) is used for modeling and predictive control of the non-linear system. The possibility of an on-line estimation of actual parameters from off-line trained neural model of the non-linear system using the gain matrix is considered in the algorithm of GPC. The neural model is linearized by means method of instantaneous linearization in each sample and an estimated parameters from neural NARX model of the non-linear system are used for design of GPC algorithm. The validity of classical and neural GPC strategy is tested by computer simulations in Matlab/Simulink language using architecture of S-functions of the library PredicLib.