Cascade complex systems — Global modeling using neural networks

In the paper global modeling of complex systems is discussed. Complex systems consists of several sub-systems connected in series. To modeling the complex system a global model is created. Thus, the global model of the complex system consists of simple models, which are connected to the corresponding structure. In the process of modeling the complex system needs a global quality criterion. As the global quality criterion was adopted a weighted sum of the quality criteria of simple models. For simple models as a quality criterion was adopted the sum of squared errors. The global quality criterion includes weighting factors that determine the participation of the quality criterion of individual simple models. Thus, the global quality criterion of the global model to a certain extent depends on the choice of the weighting factors in the global quality criterion. The investigation of influence of these weighting factors on the quality of the global model of the complex system is discussed. The investigation is examined by a complex system which is composed from three nonlinear simple plants connected in series. As the global model of the complex system multilayer feed forward neural networks are used.