Nonlinear control system using universal learning network with RBF and RasVal

Ning Shao, Kotaro Hirasawa, Masanao Ohbayashi, Junichi Murata, Kazuyuki Togo, Takayuki Furuzuki

Research output: Contribution to journalArticle

Abstract

In this paper, we present a new control method firstly for nonlinear systems using Universal Learning Network(ULN) with radial basis function(RBF). ULN can model and control the large scale complicated systems such as industrial plants, economic, social and life phenomena. The basic idea of ULN is that large scale complicated control systems can be modeled by the network which consists of nonlinearly operated nodes and branches which may have arbitrary time delays including zero or minus ones. Second, a new learning algorithm is applied to the design of the optimal network controller of a nonlinear control system. The optimization method is named RasVal, which is a kind of random searching, and it can search for a global minimum systematically and effectively in a single framework which is not a combination of different methods. The searching for a global minimum is carried out based on the probability density functions of searching, which can be modified using information on success or failure of the past searching in order to execute intensified and diversified searching. Simulation studies were carried out in the following four cases to compare the learning speed and performance: (1). comparing Radial Basis Function(RBF) with Sigmoid Function(SF) based on the gradient method. (2). comparing RBF with SF based on RasVal. (3). comparing Ras Val with the gradient method for the RBF controller. (4). comparing RasVal with the gradient method for the SF controller. By applying RasVal and the gradient method to a nonlinear crane control system, it has been proved that the simulation results of ULN with RBF based on the gradient method are superior in performance to those of neural networks, and it has also been shown that the RBF control has better performance for the generalization capability than the neural network control based on the gradient method. On the other hand, it has been shown that the neural network control based on RasVal has better performance than the RBF control. At the same time, it has been shown that the RasVal is superior in performance to the commonly used back propagation learning algorithm whether the RBF controller or the NN controller is used. On the other hand, the generalization capability of a Radial Basis Function controller using RasVal was studied and it is shown that a new method is effective to overcome the over-fitting problem in nonlinear control systems.

Original languageEnglish
Pages (from-to)197-212
Number of pages16
JournalResearch Reports on Information Science and Electrical Engineering of Kyushu University
Volume2
Issue number2
Publication statusPublished - 1997 Sep
Externally publishedYes

Fingerprint

Nonlinear control systems
Gradient methods
Controllers
Neural networks
Learning algorithms
Control systems
Cranes
Backpropagation
Probability density function
Industrial plants
Large scale systems
Nonlinear systems
Time delay

Keywords

  • Generalization capability
  • Neural network
  • Nonlinear control
  • Radial basis function
  • Random search method
  • Universal learning network

ASJC Scopus subject areas

  • Electrical and Electronic Engineering
  • Hardware and Architecture
  • Engineering (miscellaneous)

Cite this

Nonlinear control system using universal learning network with RBF and RasVal. / Shao, Ning; Hirasawa, Kotaro; Ohbayashi, Masanao; Murata, Junichi; Togo, Kazuyuki; Furuzuki, Takayuki.

In: Research Reports on Information Science and Electrical Engineering of Kyushu University, Vol. 2, No. 2, 09.1997, p. 197-212.

Research output: Contribution to journalArticle

Shao, Ning ; Hirasawa, Kotaro ; Ohbayashi, Masanao ; Murata, Junichi ; Togo, Kazuyuki ; Furuzuki, Takayuki. / Nonlinear control system using universal learning network with RBF and RasVal. In: Research Reports on Information Science and Electrical Engineering of Kyushu University. 1997 ; Vol. 2, No. 2. pp. 197-212.
@article{1fd85eb8c383424889585106d982bc3e,
title = "Nonlinear control system using universal learning network with RBF and RasVal",
abstract = "In this paper, we present a new control method firstly for nonlinear systems using Universal Learning Network(ULN) with radial basis function(RBF). ULN can model and control the large scale complicated systems such as industrial plants, economic, social and life phenomena. The basic idea of ULN is that large scale complicated control systems can be modeled by the network which consists of nonlinearly operated nodes and branches which may have arbitrary time delays including zero or minus ones. Second, a new learning algorithm is applied to the design of the optimal network controller of a nonlinear control system. The optimization method is named RasVal, which is a kind of random searching, and it can search for a global minimum systematically and effectively in a single framework which is not a combination of different methods. The searching for a global minimum is carried out based on the probability density functions of searching, which can be modified using information on success or failure of the past searching in order to execute intensified and diversified searching. Simulation studies were carried out in the following four cases to compare the learning speed and performance: (1). comparing Radial Basis Function(RBF) with Sigmoid Function(SF) based on the gradient method. (2). comparing RBF with SF based on RasVal. (3). comparing Ras Val with the gradient method for the RBF controller. (4). comparing RasVal with the gradient method for the SF controller. By applying RasVal and the gradient method to a nonlinear crane control system, it has been proved that the simulation results of ULN with RBF based on the gradient method are superior in performance to those of neural networks, and it has also been shown that the RBF control has better performance for the generalization capability than the neural network control based on the gradient method. On the other hand, it has been shown that the neural network control based on RasVal has better performance than the RBF control. At the same time, it has been shown that the RasVal is superior in performance to the commonly used back propagation learning algorithm whether the RBF controller or the NN controller is used. On the other hand, the generalization capability of a Radial Basis Function controller using RasVal was studied and it is shown that a new method is effective to overcome the over-fitting problem in nonlinear control systems.",
keywords = "Generalization capability, Neural network, Nonlinear control, Radial basis function, Random search method, Universal learning network",
author = "Ning Shao and Kotaro Hirasawa and Masanao Ohbayashi and Junichi Murata and Kazuyuki Togo and Takayuki Furuzuki",
year = "1997",
month = "9",
language = "English",
volume = "2",
pages = "197--212",
journal = "Research Reports on Information Science and Electrical Engineering of Kyushu University",
issn = "1342-3819",
publisher = "Kyushu University, Faculty of Science",
number = "2",

}

TY - JOUR

T1 - Nonlinear control system using universal learning network with RBF and RasVal

AU - Shao, Ning

AU - Hirasawa, Kotaro

AU - Ohbayashi, Masanao

AU - Murata, Junichi

AU - Togo, Kazuyuki

AU - Furuzuki, Takayuki

PY - 1997/9

Y1 - 1997/9

N2 - In this paper, we present a new control method firstly for nonlinear systems using Universal Learning Network(ULN) with radial basis function(RBF). ULN can model and control the large scale complicated systems such as industrial plants, economic, social and life phenomena. The basic idea of ULN is that large scale complicated control systems can be modeled by the network which consists of nonlinearly operated nodes and branches which may have arbitrary time delays including zero or minus ones. Second, a new learning algorithm is applied to the design of the optimal network controller of a nonlinear control system. The optimization method is named RasVal, which is a kind of random searching, and it can search for a global minimum systematically and effectively in a single framework which is not a combination of different methods. The searching for a global minimum is carried out based on the probability density functions of searching, which can be modified using information on success or failure of the past searching in order to execute intensified and diversified searching. Simulation studies were carried out in the following four cases to compare the learning speed and performance: (1). comparing Radial Basis Function(RBF) with Sigmoid Function(SF) based on the gradient method. (2). comparing RBF with SF based on RasVal. (3). comparing Ras Val with the gradient method for the RBF controller. (4). comparing RasVal with the gradient method for the SF controller. By applying RasVal and the gradient method to a nonlinear crane control system, it has been proved that the simulation results of ULN with RBF based on the gradient method are superior in performance to those of neural networks, and it has also been shown that the RBF control has better performance for the generalization capability than the neural network control based on the gradient method. On the other hand, it has been shown that the neural network control based on RasVal has better performance than the RBF control. At the same time, it has been shown that the RasVal is superior in performance to the commonly used back propagation learning algorithm whether the RBF controller or the NN controller is used. On the other hand, the generalization capability of a Radial Basis Function controller using RasVal was studied and it is shown that a new method is effective to overcome the over-fitting problem in nonlinear control systems.

AB - In this paper, we present a new control method firstly for nonlinear systems using Universal Learning Network(ULN) with radial basis function(RBF). ULN can model and control the large scale complicated systems such as industrial plants, economic, social and life phenomena. The basic idea of ULN is that large scale complicated control systems can be modeled by the network which consists of nonlinearly operated nodes and branches which may have arbitrary time delays including zero or minus ones. Second, a new learning algorithm is applied to the design of the optimal network controller of a nonlinear control system. The optimization method is named RasVal, which is a kind of random searching, and it can search for a global minimum systematically and effectively in a single framework which is not a combination of different methods. The searching for a global minimum is carried out based on the probability density functions of searching, which can be modified using information on success or failure of the past searching in order to execute intensified and diversified searching. Simulation studies were carried out in the following four cases to compare the learning speed and performance: (1). comparing Radial Basis Function(RBF) with Sigmoid Function(SF) based on the gradient method. (2). comparing RBF with SF based on RasVal. (3). comparing Ras Val with the gradient method for the RBF controller. (4). comparing RasVal with the gradient method for the SF controller. By applying RasVal and the gradient method to a nonlinear crane control system, it has been proved that the simulation results of ULN with RBF based on the gradient method are superior in performance to those of neural networks, and it has also been shown that the RBF control has better performance for the generalization capability than the neural network control based on the gradient method. On the other hand, it has been shown that the neural network control based on RasVal has better performance than the RBF control. At the same time, it has been shown that the RasVal is superior in performance to the commonly used back propagation learning algorithm whether the RBF controller or the NN controller is used. On the other hand, the generalization capability of a Radial Basis Function controller using RasVal was studied and it is shown that a new method is effective to overcome the over-fitting problem in nonlinear control systems.

KW - Generalization capability

KW - Neural network

KW - Nonlinear control

KW - Radial basis function

KW - Random search method

KW - Universal learning network

UR - http://www.scopus.com/inward/record.url?scp=0031221641&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0031221641&partnerID=8YFLogxK

M3 - Article

AN - SCOPUS:0031221641

VL - 2

SP - 197

EP - 212

JO - Research Reports on Information Science and Electrical Engineering of Kyushu University

JF - Research Reports on Information Science and Electrical Engineering of Kyushu University

SN - 1342-3819

IS - 2

ER -