Doctoral Dissertations

Keywords and Phrases

Adaptive Dynamic Programming; Event Sampled Control; Neural Network Control; Optimal Control; Q-learning

Abstract

In networked control systems (NCS), wherein a communication network is used to close the feedback loop, the transmission of feedback signals and execution of the controller is currently carried out at periodic sampling instants. Thus, this scheme requires a significant computational power and network bandwidth. In contrast, the event-based aperiodic sampling and control, which is introduced recently, appears to relieve the computational burden and high network resource utilization. Therefore, in this dissertation, a suite of novel event sampled adaptive regulation schemes in both discrete and continuous time domain for uncertain linear and nonlinear systems are designed.

Event sampled Q-learning and adaptive/neuro dynamic programming (ADP) schemes without value and policy iterations are utilized for the linear and nonlinear systems, respectively, in both the time domains. Neural networks (NN) are employed as approximators for nonlinear systems and, hence, the universal approximation property of NN in the event-sampled framework is introduced. The tuning of the parameters and the NN weights are carried out in an aperiodic manner at the event sampled instants leading to a further saving in computation when compared to traditional NN based control.

The adaptive regulator when applied on a linear NCS with time-varying network delays and packet losses shows a 30% and 56% reduction in computation and network bandwidth usage, respectively. In case of nonlinear NCS with event sampled ADP based regulator, a reduction of 27% and 66% is observed when compared to periodic sampled schemes. The sampling and transmission instants are determined through adaptive event sampling conditions derived using Lyapunov technique by viewing the closed-loop event sampled linear and nonlinear systems as switched and/or impulsive dynamical systems. "--Abstract, page iii.

Advisor(s)

Sarangapani, Jagannathan, 1965-

Committee Member(s)

Balakrishnan, S. N.
Madria, Sanjay Kumar
Zawodniok, Maciej Jan, 1975-
Chellappan, Sriram

Department(s)

Electrical and Computer Engineering

Degree Name

Ph. D. in Electrical Engineering

Sponsor(s)

National Science Foundation (U.S.)
Missouri University of Science and Technology. Intelligent Systems Center

Publisher

Missouri University of Science and Technology

Publication Date

Summer 2015

Journal article titles appearing in thesis/dissertation

  • Adaptive regulation of uncertain linear systems using Q-learning with aperiodic parameter tuning
  • Adaptive neural network based event-triggered control of single-input single-output nonlinear discrete time systems
  • Near optimal event-triggered control of nonlinear discrete-time systems using neuro dynamics programming
  • Neural network-based event-triggered state feedback control of nonlinear continuous-time systems
  • Approximate optimal control of affine nonlinear continuous time systems using event sampled neuro dynamic

Pagination

xv, 357 pages

Note about bibliography

Includes bibliographic references.

Rights

© 2015 Avimanyu Sahoo, All rights reserved.

Document Type

Dissertation - Open Access

File Type

text

Language

English

Library of Congress Subject Headings

Adaptive control systems
Dynamic programming
Neural networks (Computer science)
Linear systems
Nonlinear systems

Thesis Number

T 10766

Electronic OCLC #

921177145

Share

 
COinS