Modelling of nonlinear dynamic systems by using neural networks

This paper deals with the application of neural networks for nonlinear dynamic system modelling. It suggests some network architectures, where special static networks-networks with single trainable layer: radial basis function, cerebellar model articulation controller-and linear filters are combined in different ways. The suggested architectures can be applied successfully when some a priori information is available from the system to be modelled (gray-box modelling). Two possibilities are presented: in the first case the weights of the trainable layer are replaced by FIR filters, in the second case filtered inputs are applied to the network. Both versions can be applied in feedforward or feedback structures. The paper deals with the modelling capabilities of these architectures and derives the training equations. The capabilities and the limitations of the suggested networks are illustrated by simulation results