Abstract
This paper discusses a design method of “neural controller” which achieves the model following control for a SISO nonlinear discrete-time system. If the system dynamics is completely known, the model following control law can be derived as a nonlinear function of input-output sequence via the nonlinear control theory. Under reasonable assumptions, a neural network CMAC-Cerebellar Model Arithmetic Computer-is able to learn both the uncertain nonlinear dynamics of the system and the nonlinear function which yields the model following controller. Moreover, in order to improve the accuracy and the speed of learning, the data presented to the CMAC are transformed by the difference method or the correlation method. It is shown by computer simulations that our method works well and the correlation method is better than the difference method.
