Summary: | 博士 === 國立中央大學 === 電機工程研究所 === 83 === Virtually all dynamical systems are in some way disturbed by
uncontrolled external forces. This disturbances may assume a
myriad of forms such as wind gusts, gravity gradients, the-
rmal gradients, structural vibrations, or sensor and actuator
noise, and may enter the system in many different ways. For the
theory of control systems, the disturbances are often mo- deled
as a Gaussian noise process, and the pointing specific- ations
are expressed in terms of allowable Root-Mean-Squared (RMS)
pointing error. As a general rule, the control engineer first
designs a controller and then computes the covariance matrix of
the closed-loop system to assess the state or output RMS
response. The major difficulty encountered here is that one is
attempting to indirectly satisfy very specific control
objectives. Moreover, the requirements are multiobjective si-
nce RMS constraints are imposed on many outputs or states.
Certainly a direct design method for complete variance (RMS
squared) assignment would be useful. A complete approach, which
is called covariance control theory, for assigning the entire
state covariance matrix to the closed-loop systems via state
feedback or output dynamic controllers has been developed in
past years. However, the systems considered in these
literatures are often linear ones. Hence, in this dissertation
we will first attempt to extend the covariance control theory
to other different stochastic systems. Furthermore, the second
purpose of this dissertation is to describe how the covariance
controller design problem can be solved for simultaneously
achieving multiobjective pe- rformance requirements for linear
stochastic systems. These performance constraints considered in
this dissertation incl- ude circular closed-loop system pole
constraints, individual state variance constraints, H∞ norm
and minimum auxiliary entropy constraints of closed-loop system
transfer function.
|