Modelling of nonlinear systems by feedforward and recurrent neural networks

Abstract

Two types of artificial neural networks are studied in this paper in modelling nonlinear dynamical systems: a feedforward neural network and a recurrent neural network. When the feedforward network is used to model a dynamical system, the inputs to network include the past inputs and outputs of the plant in addition to the present input to the plant. Suitable number of past inputs and outputs depends on the assumption on model structure. For the recurrent network with a hybrid (feedforward and feedback) structure, explicit use of past inputs and outputs is not necessary for modelling since their effects are captured by the network internal states. Simulation results clearly illustrate the difference between the capability of the two networks in detecting system structures which are implicitly contained in the input-output data.