| Summary: | 博士 === 國立成功大學 === 工程科學系 === 102 === System identification is the problem of building a dynamic system model from designed input and measured output data. Usually such an approach yields accurate-enough models and is often used by control area for the on/off-line applications. Methods for realizing linear time invariant (LTI) models of dynamical systems have been developed, but real world phenomena are always nonlinear in nature. Therefore, much work remains to be done to realize nonlinear models that represent the majority of physical systems.
A special class, bilinear systems, exist as a gateway between the worlds of linear systems and of nonlinear systems. A bilinear plant model constituting a state and control-input coupling term in addition to a linear term forms the simplest model of nonlinear systems. In spite of their simple form, bilinear systems are widely applied to engineering, biology and economics. Indeed, the continuous-time bilinear system model is a universal approximater for a class of nonlinear systems and bi-linearizion procedures for nonlinear systems has widen the range of application of bilinear systems.
In 2005, Juang presented a method using a combination of pulses via multiple experiments to identify the bilinear state-space matrices exactly for noise-free data, but multiple experiments usually consume money and time. Based on realization problem of LTI systems, the basic structure theory of continuous-time bilinear systems, and the above mentioned Juang's research results, this dissertation presents a series of methods to identify bilinear systems by a single experiment, which significantly advances the original version toward the real-time/on-line applications.
|
|---|
