MODELING USING REGULARITY CRITERION BASED CONSTRUCTED NEURAL NETWORKS

Models of dynamical systems are used for many purposes like control, p rediction, simulation, filter design, reconstruction, etc. System iden tification can be divided into structure and parameter identification. However, in most system identification approaches the structure is pr esumed; only a parameter identification is performed to determine the coefficients of the functional system. Thus, if there is little knowle dge about the system structure, those approaches are not very effectiv e. In this paper a black-box input variable identification based on th e regularity criterion in GMDH (group method of data handling) using a feedforward neural network is discussed.