Masters Theses
Abstract
"Dynamic Programming is an exact method of determining optimal control for a discretized system. Unfortunately, for nonlinear systems the computations necessary with this method become prohibitive.
This project investigates the use of adaptive neural networks that utilize dynamic programming methodology to develop near optimal control laws. First, a one dimensional infinite horizon problem is examined. The first part of this thesis considers problems involving cost functions with final state constraints for one dimensional linear and nonlinear systems. It also investigates a two dimensional linear problem. In addition to these examples, an example of the corrective capabilities of critics is shown. The study shows a method of reducing the number of neural networks necessary using this method. It also mentions telescoping and normalization, two topics which may aid in neural network convergence.
The second portion of the thesis examines higher order systems. First, a four-dimensional finite horizon problem is examined. Next, an infinite horizon four-dimensional linear problem, that of an inverted pendulum in a cart, is solved. Finally, the adaptive critic method is used to find the optimal control policy for the nonlinear equations in the inverted pendulum problem.
This thesis is intended to provide a springboard to the use of adaptive critics and neural networks to control high ordered nonlinear systems in an optimal fashion"-- Abstract, p. iv
Advisor(s)
Balakrishnan, S. K.
Committee Member(s)
Rao, Vittal S.
Krishnamurthy, K.
Department(s)
Mechanical and Aerospace Engineering
Degree Name
M.S. in Mechanical Engineering
Publisher
University of Missouri--Rolla
Publication Date
Fall 1994
Pagination
ix, 69 pages
Note about bibliography
Includes bibliographical references (pages 54-68)
Rights
© 1994 Victor Lynn Biega, All rights reserved.
Document Type
Thesis - Open Access
File Type
text
Language
English
Thesis Number
T 6850
Print OCLC #
32770285
Recommended Citation
Biega, Victor, "Adaptive critic based neural networks for control" (1994). Masters Theses. 1378.
https://scholarsmine.mst.edu/masters_theses/1378