Login | Register

Identification and Control of Nonlinear Singularly Perturbed Systems Using Multi-time-scale Neural Networks


Identification and Control of Nonlinear Singularly Perturbed Systems Using Multi-time-scale Neural Networks

Zheng, Dongdong (2017) Identification and Control of Nonlinear Singularly Perturbed Systems Using Multi-time-scale Neural Networks. PhD thesis, Concordia University.

Text (application/pdf)
Zheng_PhD_S2017.pdf - Accepted Version
Available under License Spectrum Terms of Access.


Many industrial systems are nonlinear with "slow" and "fast" dynamics because of the presence of some ``parasitic" parameters such as small time constants, resistances, inductances, capacitances, masses and moments of inertia. These systems are usually labeled as "singularly perturbed" or ``multi-time-scale" systems. Singular perturbation theory has been proved to be a useful tool to control and analyze singularly perturbed systems if the full knowledge of the system model parameters is available. However, the accurate and faithful mathematical models of those systems are usually difficult to obtain due to the uncertainties and nonlinearities.

To obtain the accurate system models, in this research, a new identification scheme for the discrete time nonlinear singularly perturbed systems using multi-time-scale neural network and optimal bounded ellipsoid method is proposed firstly. Compared with other gradient descent based identification schemes, the new identification method proposed in this research can achieve faster convergence and higher accuracy due to the adaptively adjusted learning gain. Later, the optimal bounded ellipsoid based identification method for discrete time systems is extended to the identification of continuous singularly perturbed systems. Subsequently, by adding two additional terms in the weight's updating laws, a modified identification scheme is proposed to guarantee the effectiveness of the identification algorithm during the whole identification process. Lastly, through introducing some filtered variables, a robust neural network training algorithm is proposed for the system identification problem subjected to measurement noises.

Based on the identification results, the singular perturbation theory is introduced to decompose a high order multi-time-scale system into two low order subsystems -- the reduced slow subsystem and the reduced fast subsystem. Then, two controllers are designed for the two subsystems separately. By using the singular perturbation theory, an adaptive controller for a regulation problem is designed in this research firstly. Because the system order is reduced, the adaptive controller proposed in this research has a simpler structure and requires much less computational resources, compared with other conventional controllers. Afterward, an indirect adaptive controller is proposed for solving the trajectory tracking problem. The stability of both identification and control schemes are analyzed through the Lyapunov approach, and the effectiveness of the identification and control algorithms are demonstrated using simulations and experiments.

Divisions:Concordia University > Gina Cody School of Engineering and Computer Science > Mechanical and Industrial Engineering
Item Type:Thesis (PhD)
Authors:Zheng, Dongdong
Institution:Concordia University
Degree Name:Ph. D.
Program:Mechanical Engineering
Date:16 January 2017
Thesis Supervisor(s):Xie, Wen-Fang
ID Code:982124
Deposited On:01 Jun 2017 12:51
Last Modified:18 Jan 2018 17:54
All items in Spectrum are protected by copyright, with all rights reserved. The use of items is governed by Spectrum's terms of access.

Repository Staff Only: item control page

Downloads per month over past year

Back to top Back to top