Model reference based neural network adaptive controller

Linear system theory has had significant contributions to developments in the area of classical controls in the past three decades. The motivation of this work emerges from the need to develop novel control strategies that can be applied to nonlinear dynamic systems. Furthermore, the need for an adaptive scheme emerges for dealing with time varying systems. This paper presents model reference based neural network structure that can be used for adaptive control of linear and nonlinear processes, The proposed neural network controller is tested on several simulated nonlinear systems. Also, a fast algorithm is introduced for training the proposed neural network controller. This algorithm is based on Davidon's least squares minimization technique. Finally, a neural network linearization methodology is presented that provides a framework under which the local stability of the feedback control system can be analyzed. (C) 1998 Elsevier Science Ltd. All rights reserved.