Nonlinear System Approximation Using Dynamic Neural Networks

Abstract

Identification of dynamic systems is often an important prerequisite for a successful analysis and controller design. Due to the nonlinear nature of most of the processes encountered in engineering applications, there has been extensive research covering the field of nonlinear system identification [109, 110, 111, 112]. It is here that the use of neural networks emerges as a feasible solution. The universal approximation properties of static neural networks [77] make them a useful tool for modelling nonlinear systems. The problem of nonlinear modelling using static neural networks has been extensively researched [33, 113] and many approaches have used multilayer perceptrons and radial basis functions [110, 114, 115].