Smart neural control of pure-feedback systems

Abstract

In this paper, by combining smart neural design with a recently proposed ISS-modular neural control approach, we present a smart neural control scheme for general (non-affine) pure-feedback systems. Although the neural controller in achieves a semi-global result for general (non-affine) pure-feedback systems, it is by nature a high-order dynamic controller, which cannot be reduced in general due to its need of simultaneous adaptation of a large number of neural weights. To overcome this problem, in this paper we develop a smart neural controller, which on the contrary is a static and low-order controller, hence more computationally feasible in practical design and implementation. To improve the NN generalization ability, which plays an important role in our smart neural control scheme, chaotic reference signals are employed in the training phase of the scheme, where the complex chaotic signals offer much richer information for NN learning due to the ergodicity of chaos. Since pure-feedback system represents a very large class of nonlinear systems, the smart neural control scheme is expected to be useful for a wide variety of industrial applications.