Global models of dynamic complex systems - modelling using the multilayer neural networks

In this paper, global models of dynamic complex systems using the neural networks is discussed. The description of a complex system is given by a description of each system element and structure. As a model the multilayer neural networks with the tapped delay line (TDL), which have the same structure as a complex system, are accepted. Two approaches, a global model and a global model with the quality local model taken into account are proposed. To learn global models the modified back-propagation algorithms have been developed for the unique structure of the complex model. To model dynamic simple plants, of which the complex system is composed, a series-parallel model of identification using the feedforward network with the tapped delay line (TDL) and the feedback loops, in which the gradient can be calculated by means of the simpler static back-propagation method is proposed. Computer simulations were performed for the dynamic complex system, which consists of two dynamic nonlinear simple plants connected in series, described by means of nonlinear difference equations.