Identification of ball and plate system using multiple neural network models

Based on the dynamic characteristics of the ball and plate system, the mathematic description of system is usually derived by using Lagrange equations, but the controller of the system is very difficult to design according to this kind of model. For the convenience of controller design, NARMA and NARMA-L2 model are always used to describe the nonlinear system. Two kinds of nonlinear models (NARMA and NARMA-L2 model) are set up for the ball and plate system, and BP neural network will be trained to approximate the nonlinear function of these two models. The comparison of different models is made by simulations, and the advantage and shortcoming of different models are analyzed in details. The results of this paper will be very useful for the study of modeling and controller design of complex nonlinear system.