Identification of Non-linear Structures Using Recurrent Neural Networks

Two different partially recurrent neural networks st,ructured as Multi Layer Perceptrons (MLP) are inves tigated for time domain identification of a nonlinear structure. The one partially recurrent neural network has feedback of a displacement component from t,he output layer to a tapped-delay-line (TDL) input layer. The other recurrent neural network bazd on the Innovation State Space model (ISSM) has feedback of the state space vector from the output layer to the input layer. The recurrent neural network approaches are validated with respect to prediction and simulation of a nonlinear process by application to simulated data from a viscous damped oscillator with hyst.eresis of the curveliniar type described hy the Bow-Wen model. The oscillator is subjected to amplitude modulated Gaussian white noise filtered through a Kanai-Tajimi filter. It is found that the two neural network models can act as actual system identifiers, predictors and simulators. The recurrent neural network with a TDL seems to he a better simulator than the ISSM n&work.