東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

STABILITY AND CONVERGENCE OF PARALLE...

KREUTZ, KENNETH KEITH.

Linked to FindBook

Google Book

Amazon

博客來

STABILITY AND CONVERGENCE OF PARALLEL ADAPTIVE REGULATORS: TOWARDS A ROBUST PERSPECTIVE (IDENTIFICATION, UNCERTAIN, CONTROL, BAYESIAN, LEARNING).

Record Type:	Language materials, printed : Monograph/item
Title/Author:	STABILITY AND CONVERGENCE OF PARALLEL ADAPTIVE REGULATORS: TOWARDS A ROBUST PERSPECTIVE (IDENTIFICATION, UNCERTAIN, CONTROL, BAYESIAN, LEARNING)./
Author:	KREUTZ, KENNETH KEITH.
Description:	267 p.
Notes:	Source: Dissertation Abstracts International, Volume: 46-02, Section: B, page: 0624.
Contained By:	Dissertation Abstracts International46-02B.
Subject:	Engineering, System Science. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=8509091

STABILITY AND CONVERGENCE OF PARALLEL ADAPTIVE REGULATORS: TOWARDS A ROBUST PERSPECTIVE (IDENTIFICATION, UNCERTAIN, CONTROL, BAYESIAN, LEARNING).
KREUTZ, KENNETH KEITH.

STABILITY AND CONVERGENCE OF PARALLEL ADAPTIVE REGULATORS: TOWARDS A ROBUST PERSPECTIVE (IDENTIFICATION, UNCERTAIN, CONTROL, BAYESIAN, LEARNING). - 267 p.

Source: Dissertation Abstracts International, Volume: 46-02, Section: B, page: 0624.

Thesis (Ph.D.)--University of California, San Diego, 1985.

Parallel adaptive regulators, such as the Multiple Model Adaptive Controller, have long been promoted as a practical way to regulate uncertain systems, primarily because of the belief that such regulators will learn a correct model via convergence of a posteriori probabilities. In this dissertation a theoretical framework is developed which provides an understanding of such regulators give stable and/or learning control. The issues of convergence, i.e. learning, and of stability are separated since if learning can be shown to occur under realistic and possibly unstable closed-loop operating conditions, a convergence-based stability analysis is possible.Subjects--Topical Terms:

1018128
Engineering, System Science.

STABILITY AND CONVERGENCE OF PARALLEL ADAPTIVE REGULATORS: TOWARDS A ROBUST PERSPECTIVE (IDENTIFICATION, UNCERTAIN, CONTROL, BAYESIAN, LEARNING).
LDR:03103nam 2200253 a 45 001 929847
005 20110427
008 110427s1985 d
035 $a (UnM)AAI8509091
035 $a AAI8509091
040 $a UnM $c UnM
100 1 $a KREUTZ, KENNETH KEITH. $3 1253333
245 1 0 $a STABILITY AND CONVERGENCE OF PARALLEL ADAPTIVE REGULATORS: TOWARDS A ROBUST PERSPECTIVE (IDENTIFICATION, UNCERTAIN, CONTROL, BAYESIAN, LEARNING).
300 $a 267 p.
500 $a Source: Dissertation Abstracts International, Volume: 46-02, Section: B, page: 0624.
502 $a Thesis (Ph.D.)--University of California, San Diego, 1985.
520 $a Parallel adaptive regulators, such as the Multiple Model Adaptive Controller, have long been promoted as a practical way to regulate uncertain systems, primarily because of the belief that such regulators will learn a correct model via convergence of a posteriori probabilities. In this dissertation a theoretical framework is developed which provides an understanding of such regulators give stable and/or learning control. The issues of convergence, i.e. learning, and of stability are separated since if learning can be shown to occur under realistic and possibly unstable closed-loop operating conditions, a convergence-based stability analysis is possible.
520 $a Generally the assumptions which are made to enable construction of posterior probabilities are in error and one is concerned with the convergence of pseudo probabilities. Such convergence is an important issue in adaptive (Bayesian) estimation. Using a Prediction Error approach and the Martingale Convergence Theorem as pioneered by Ljung, it is shown that appropriately constructed pseudo probabilities can converge to pick out a model even when signals can grow without bound and even when no model accurately describes the plant of interest. This work can be considered a logical extension of that of Anderson, Moore, and Hawkes particularly since the need for finite power signals is removed.
520 $a The model learned by the pseudo probabilities considered here generally provides smallest one-step ahead prediction errors. Asking when such a learned model is appropriate for regulator design, answers are given for some important special cases. A robustness perspective based on the work of Doyle, Stein, and others is the key to constructing a framework within which parallel adaptive controllers are well posed when every model gives an incorrect plant description. Assuming the existence of a stabilizing (not error-free) model, it is shown that if this model is learned, then stability of parallel adaptive regulators follows. An example is given which shows the effectiveness of our approach to understanding convergence and stability issues and which demonstrates the need for a robustness based approach both in the choice of the admissible models and in LQG regulator design.
590 $a School code: 0033.
650 4 $a Engineering, System Science. $3 1018128
690 $a 0790
710 2 0 $a University of California, San Diego. $3 1018093
773 0 $t Dissertation Abstracts International $g 46-02B.
790 $a 0033
791 $a Ph.D.
792 $a 1985
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=8509091