A neuro-predictive based self-tuning controller

A self-tuning controller for a class of processes with predictable dynamics variations using model predictive scheme is presented in this paper. The self-tuning controller design is based on the optimization of a cost function subject to constraints over a finite prediction horizon in time, and, it uses a neural process model. The performance of this new self-tuning controller is substantiated by experiments on a pH control system. Simulation results show how the model predictive scheme is involved in the self-tuning control of nonlinear systems.

I. Introduction

Classical control systems design for complex processes involves complex mathematical analysis and the resulting control architectures have difficulties in controlling highly nonlinear plants. To avoid these difficulties, new approaches based on neural networks for control have been proposed in the last years. The use of neural networks'learning ability makes controller design to be flexible when plant dynamics are complex and highly nonlinear. The use of dynamical neural networks for identification and control was first introduced in [1] and bridging the gap between theory and applications, as well as the relation between neural and adaptive control, have been pointed out in [2]. A systematic classification of neural networks-based control systems can be found in [3] and [4].