Intelligent control using neural networks and multiple models

In this paper a new framework for intelligent control is established to adaptively control a class of nonlinear discrete time dynamical systems while assuring boundedness of all signals. A linear robust adaptive controller and multiple nonlinear neural network based adaptive controllers are used, and a switching law is suitably defined to switch between them, based upon their performance in predicting the plant output. Boundedness of all the signals is established regardless of the parameter adjustment mechanism of the neural network controllers, and thus neural network models can be used in novel ways to better detect changes in the system and provide starting points for adaptation. The effectiveness of the proposed approach is demonstrated by simulation studies.