ニューラルネットワークを用いたウィナーシステムの同定と一般化予測制御

概要

論文の詳細を見る
The Wiener model consists of a linear dynamic block in cascade with a static nonlinearity. And many physical systems can be naturally described by the Wiener systems. The linear subsystem is modeled by the transfer function model, and the non-linear function of the Wiener system is expressed by the artificial neural networks which have the ability to learn complex nonlinear relationships. The parameters of both linear and non-linear subsystems are estimated simultaneously, by optimizing the nonlinear objective function which is the total error of the system and the model.As the order of linear subsystem and the adequate number of hidden neurons are unknown, they are estimated by the minimum description length criteria. The generalized predictive control is applied to the linear subsystem, and the estimated parameters of linear and non-linear subsystem are used adaptively to the generalized predictive control. The numerical examples demonstrate the validity of both proposed identification method and control design.
社団法人電気学会の論文
2003-04-01