Practical methods of tracking of nonstationary time series applied to real-world data

Ian T. Nabney, Alan McLachlan, David Lowe

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In this paper, we discuss some practical implications for implementing adaptable network algorithms applied to non-stationary time series problems. Two real world data sets, containing electricity load demands and foreign exchange market prices, are used to test several different methods, ranging from linear models with fixed parameters, to non-linear models which adapt both parameters and model order on-line. Training with the extended Kalman filter, we demonstrate that the dynamic model-order increment procedure of the resource allocating RBF network (RAN) is highly sensitive to the parameters of the novelty criterion. We investigate the use of system noise for increasing the plasticity of the Kalman filter training algorithm, and discuss the consequences for on-line model order selection. The results of our experiments show that there are advantages to be gained in tracking real world non-stationary data through the use of more complex adaptive models.
Original languageEnglish
Title of host publicationApplications and science of artificial neural networks II
EditorsS.K. Rogers, D.W. Ruck
PublisherSPIE
Pages152-163
Number of pages12
Volume2760
DOIs
Publication statusPublished - 9 Apr 1996
EventAeroSense '96 : Applications and Science of Artificial Neural Networks II - Orlando, United States
Duration: 8 Apr 199612 Apr 1996

Publication series

NameSPIE proceedings
PublisherSPIE
Volume2760
ISSN (Print)0277-786X

Conference

ConferenceAeroSense '96 : Applications and Science of Artificial Neural Networks II
CountryUnited States
CityOrlando
Period8/04/9612/04/96

Fingerprint

Time series
Radial basis function networks
Extended Kalman filters
Kalman filters
Plasticity
Dynamic models
Electricity
Experiments

Bibliographical note

Ian T. Nabney ; Alan McLachlan and David Lowe, "Practical methods of tracking of nonstationary time series applied to real-world data", Proc. SPIE 2760, Applications and Science of Artificial Neural Networks II, 152 (March 22, 1996); doi:10.1117/12.235906.

Copyright 1996 Society of Photo-Optical Instrumentation Engineers. One print or electronic copy may be made for personal use only. Systematic reproduction and distribution, duplication of any material in this paper for a fee or for commercial purposes, or modification of the content of the paper are prohibited.

http://dx.doi.org/10.1117/12.235906

Keywords

  • RBF network
  • RAN
  • noise
  • plasticity
  • Kalman filter
  • training algorithm
  • tracking
  • real world non-stationary
  • complex adaptive models

Cite this

Nabney, I. T., McLachlan, A., & Lowe, D. (1996). Practical methods of tracking of nonstationary time series applied to real-world data. In S. K. Rogers, & D. W. Ruck (Eds.), Applications and science of artificial neural networks II (Vol. 2760, pp. 152-163). (SPIE proceedings; Vol. 2760). SPIE. https://doi.org/10.1117/12.235906
Nabney, Ian T. ; McLachlan, Alan ; Lowe, David. / Practical methods of tracking of nonstationary time series applied to real-world data. Applications and science of artificial neural networks II. editor / S.K. Rogers ; D.W. Ruck. Vol. 2760 SPIE, 1996. pp. 152-163 (SPIE proceedings).
@inproceedings{3c8fde0017ec44e0b354518134899d5e,
title = "Practical methods of tracking of nonstationary time series applied to real-world data",
abstract = "In this paper, we discuss some practical implications for implementing adaptable network algorithms applied to non-stationary time series problems. Two real world data sets, containing electricity load demands and foreign exchange market prices, are used to test several different methods, ranging from linear models with fixed parameters, to non-linear models which adapt both parameters and model order on-line. Training with the extended Kalman filter, we demonstrate that the dynamic model-order increment procedure of the resource allocating RBF network (RAN) is highly sensitive to the parameters of the novelty criterion. We investigate the use of system noise for increasing the plasticity of the Kalman filter training algorithm, and discuss the consequences for on-line model order selection. The results of our experiments show that there are advantages to be gained in tracking real world non-stationary data through the use of more complex adaptive models.",
keywords = "RBF network, RAN, noise, plasticity, Kalman filter, training algorithm, tracking, real world non-stationary, complex adaptive models",
author = "Nabney, {Ian T.} and Alan McLachlan and David Lowe",
note = "Ian T. Nabney ; Alan McLachlan and David Lowe, {"}Practical methods of tracking of nonstationary time series applied to real-world data{"}, Proc. SPIE 2760, Applications and Science of Artificial Neural Networks II, 152 (March 22, 1996); doi:10.1117/12.235906. Copyright 1996 Society of Photo-Optical Instrumentation Engineers. One print or electronic copy may be made for personal use only. Systematic reproduction and distribution, duplication of any material in this paper for a fee or for commercial purposes, or modification of the content of the paper are prohibited. http://dx.doi.org/10.1117/12.235906",
year = "1996",
month = "4",
day = "9",
doi = "10.1117/12.235906",
language = "English",
volume = "2760",
series = "SPIE proceedings",
publisher = "SPIE",
pages = "152--163",
editor = "S.K. Rogers and D.W. Ruck",
booktitle = "Applications and science of artificial neural networks II",
address = "United States",

}

Nabney, IT, McLachlan, A & Lowe, D 1996, Practical methods of tracking of nonstationary time series applied to real-world data. in SK Rogers & DW Ruck (eds), Applications and science of artificial neural networks II. vol. 2760, SPIE proceedings, vol. 2760, SPIE, pp. 152-163, AeroSense '96 : Applications and Science of Artificial Neural Networks II, Orlando, United States, 8/04/96. https://doi.org/10.1117/12.235906

Practical methods of tracking of nonstationary time series applied to real-world data. / Nabney, Ian T.; McLachlan, Alan; Lowe, David.

Applications and science of artificial neural networks II. ed. / S.K. Rogers; D.W. Ruck. Vol. 2760 SPIE, 1996. p. 152-163 (SPIE proceedings; Vol. 2760).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Practical methods of tracking of nonstationary time series applied to real-world data

AU - Nabney, Ian T.

AU - McLachlan, Alan

AU - Lowe, David

N1 - Ian T. Nabney ; Alan McLachlan and David Lowe, "Practical methods of tracking of nonstationary time series applied to real-world data", Proc. SPIE 2760, Applications and Science of Artificial Neural Networks II, 152 (March 22, 1996); doi:10.1117/12.235906. Copyright 1996 Society of Photo-Optical Instrumentation Engineers. One print or electronic copy may be made for personal use only. Systematic reproduction and distribution, duplication of any material in this paper for a fee or for commercial purposes, or modification of the content of the paper are prohibited. http://dx.doi.org/10.1117/12.235906

PY - 1996/4/9

Y1 - 1996/4/9

N2 - In this paper, we discuss some practical implications for implementing adaptable network algorithms applied to non-stationary time series problems. Two real world data sets, containing electricity load demands and foreign exchange market prices, are used to test several different methods, ranging from linear models with fixed parameters, to non-linear models which adapt both parameters and model order on-line. Training with the extended Kalman filter, we demonstrate that the dynamic model-order increment procedure of the resource allocating RBF network (RAN) is highly sensitive to the parameters of the novelty criterion. We investigate the use of system noise for increasing the plasticity of the Kalman filter training algorithm, and discuss the consequences for on-line model order selection. The results of our experiments show that there are advantages to be gained in tracking real world non-stationary data through the use of more complex adaptive models.

AB - In this paper, we discuss some practical implications for implementing adaptable network algorithms applied to non-stationary time series problems. Two real world data sets, containing electricity load demands and foreign exchange market prices, are used to test several different methods, ranging from linear models with fixed parameters, to non-linear models which adapt both parameters and model order on-line. Training with the extended Kalman filter, we demonstrate that the dynamic model-order increment procedure of the resource allocating RBF network (RAN) is highly sensitive to the parameters of the novelty criterion. We investigate the use of system noise for increasing the plasticity of the Kalman filter training algorithm, and discuss the consequences for on-line model order selection. The results of our experiments show that there are advantages to be gained in tracking real world non-stationary data through the use of more complex adaptive models.

KW - RBF network

KW - RAN

KW - noise

KW - plasticity

KW - Kalman filter

KW - training algorithm

KW - tracking

KW - real world non-stationary

KW - complex adaptive models

UR - http://proceedings.spiedigitallibrary.org/proceeding.aspx?articleid=1019119

UR - http://www.scopus.com/inward/record.url?scp=0029721642&partnerID=8YFLogxK

U2 - 10.1117/12.235906

DO - 10.1117/12.235906

M3 - Conference contribution

VL - 2760

T3 - SPIE proceedings

SP - 152

EP - 163

BT - Applications and science of artificial neural networks II

A2 - Rogers, S.K.

A2 - Ruck, D.W.

PB - SPIE

ER -

Nabney IT, McLachlan A, Lowe D. Practical methods of tracking of nonstationary time series applied to real-world data. In Rogers SK, Ruck DW, editors, Applications and science of artificial neural networks II. Vol. 2760. SPIE. 1996. p. 152-163. (SPIE proceedings). https://doi.org/10.1117/12.235906