Smart neural control of uncertain non-linear systems

In this paper, we present a 'smart' neural control scheme for uncertain non-linear systems using the localized radical basis function (RBF) networks. This scheme is designed such that the current control action can utilize the knowledge that the NN learned from the past control process. Compared with most existing adaptive neural controllers, which are in general very-high-order dynamic controllers due to the simultaneous adaptation of a large number of neural weights, the smart RBF neural controller is a static and low-order one, and thus is more computationally feasible in practical design and implementation. To improve the generalization ability of the RBF networks, which plays an important role in the smart neural control scheme, chaotic reference signals are employed in the training phase of the scheme, where the complex chaotic signals offer richer information for NN learning due to the ergodicity of chaos. The proposed neural control scheme can act 'smartly' in the operational phase after the RBF networks have been well-trained in the training phase, in a way similar to the process that humans accomplish some complicated control tasks easily after the 'neuro-controllers' in their brains have been well-trained previously. The smart neural control scheme also provides a strong motivation for the current research on chaos generation. Simulation studies are included to demonstrate the effectiveness of the new control scheme. Copyright © 2003 John Wiley & Sons, Ltd.