Study the Quality of Global Neural Model with Regard to the Local Models of Dynamic Complex System

Abstract

In this paper a dynamic neural global model with regard to local models of dynamic complex systems is discussed. A dynamic complex system which consists of two nonlinear discrete time sub-systems is considered. As a global model with regard to local models multilayer neural networks in a dynamic structure with TDL are used. The global model with regard local models is composed of two simple models and two local models. One way to take account of local models in the global model is an appropriate definition of a synthetic quality criterion. The criterion of the whole model. That criterion contains coefficients which define the participation of the quality criteria of the global model and local models. The contribution of this work is study of the influence these coefficients on the global model quality. The influence is examined for a special learning algorithm for dynamic complex neural networks.