Universal neural controllers and stability analysis

Abstract

A set Γ of neural controllers is a universal neural controller if and only if any given controllable process P can also be controlled by a specific neural controller in Γ. Given any continuous real-valued controller on a compact set, there exists an element in the set Γ of PD/PI type inverse gaussian and gaussian potential function neural controllers (IGFNCs, GPFNCs) that could approximate the given continuous controller to any degree of accuracy. An application of this result shows that this type of radial basis function neural controller (RBFNCs) is a universal neural controller. Hence, any given controllable process can be controlled by this type of RBFNCs operating simultaneously. To realize the universal neural controller in practical processes, a stable region of PD type GPFNC control system is studied in parametric space of the gaussian potential frunctton network (GPFN)