HIGH-ORDER NEURAL-NETWORK STRUCTURES FOR IDENTIFICATION OF DYNAMICAL-SYSTEMS

Several continuous-time and discrete-time recurrent neural network mod els have been developed and-applied to various engineering problems. O ne of the difficulties encountered in the application of recurrent net works is the derivation of efficient learning algorithms that also gua rantee stability of the overall system. This paper studies the approxi mation and learning properties of one class of recurrent networks, kno wn as high-order neural networks, and applies these architectures to t he identification of dynamical systems. In recurrent high-order neural networks the dynamic components are distributed throughout the networ k in the form of dynamic neurons. It is shown that if enough high-orde r connections are allowed then this network is capable of approximatin g arbitrary dynamical systems. Identification schemes based on high-or der network architectures are designed and analyzed.