Nonlinear System Modeling using Neural Networks

Artificial neural networks have gained increasing popularity in control area in recent years. This paper outlines the application of a neural network based identification approach to a dynamical nonlinear system, a cantilever plate with distributed actuators and sensors. The type of neural networks utilized are multi-layer perceptrons with the backpropagation (BP) learning method. The identifier is implemented in discrete-time domain, and its performance is compared with a linear model from a previous result, that used frequency domain method. The time-domain neural network approach displays better nonlinear dynamical properties. A new efficient scheme to train the BP neural networks with a large amount of data is also introduced.