Adaptive neural network control schemes for unknown nonlinear dynamic systems

This paper investigates the applicability of neural networks for unknown nonlinear dynamic systems. Traditionally the design of an adaptive controller of a nonlinear system starts with parametric estimation model whose functional form is known; whether it is an analytical model or a regression model. In this paper, a neural network control paradigm consisting of a neural network controller and neural network on-line system identifier is presented. The adaptive inverse model (AIM) neural network control scheme assumes no knowledge about the functional form of the process. One implementation uses feed forward networks and the other one uses cerebellar model articulation controller (CMAC). The proposed scheme is applied to a simple nonlinear control example for verification and comparison.