ABSTRACT
This chapter presents the structure and properties of Group Method of Data Handling (GMDH) neural networks. It outlines the fundamentals of GMDH neural networks and presents the processes of the synthesis of structure and parameters estimation of GMDH network. The chapter considers problem of the modelling of dynamical systems and describes the processes of the synthesis of single-input and multi-output GMDH neural networks. It shows an application of the GMDH neural model to the robust fault detection. The concept of the GMDH approach relies on replacing the complex neural model by the set of hierarchically connected partial models. The application of the GMDH approach during neural network synthesis allows to apply parameter estimation of linear-in-parameters model algorithms, for example, the least mean square. Thus, during system identification, it seems desirable to employ models, which can represent the dynamics of the system.
