CHAOTIC SYSTEM IDENTIFICATION BY NEURAL NETWORKS AND ITS APPLICATION TO VIBRATION CONTROL

摘要

This research develops the identification of chaotic dynamical system using neural networks and its application to vibration control. In conventional control theory, it is difficult to identify a mathematical model of practical system because of system complexity and existence of nonlinearity. Instead of theoretical method, neural networks with different architecture have been applied to the identification and control for wide class of nonlinear systems. In this research, we propose a learning control scheme to realize the identification and control of chaotic vibratory system, using multi-layered neural networks, in which the control performance is satisfied for an unknown controlled object by repeated trials. First, the effectiveness of neural networks is shown in the identification problem of Buffing's chaotic time series data and the possibility of short-term predictability in chaotic vibratory system discussed. Second, a new technique is proposed which identifies the nonlinear systems as a nonlinear mapping in forward direction and employs its inverse mapping as a controller. As a result, it is shown that identification result of chaotic time series data is available for vibratory control system.