Comparative analysis of Gaussian Process power curve models based on different stationary covariance functions for the purpose of improving model accuracy

Research output: Contribution to journalArticle

3 Citations (Scopus)

Abstract

Gaussian Process (GP) models are increasingly finding application in wind turbine condition monitoring and in particular early fault detection. GP model accuracy is greatly influenced by the choice and type of the covariance functions (used to described the similarity between two given data points). Hence, the appropriate selection and composition of covariance functions is essential for accurate GP modelling.
In this study, an in-depth analysis of commonly used stationary covariance functions is presented in which wind turbine power curve used where GP based power curve has been constructed using different stationary covariance functions, and after that, a comparative analysis has been carried out in order to identify the most effective covariance function. The commonly used squared exponential covariance function is taken as the benchmark, against which other covariance functions are assessed.
The results show that the performance (in terms of model accuracy and uncertainty) of GP fitted power curve models based on rational quadratic covariance functions is almost the same as for the most commonly used squared exponential function. Thus, rational quadratic covariance functions can be used instead of squared exponential covariance functions. In this paper, strength and weakness of stationary covariance functions would be highlighted for effective condition monitoring.
LanguageEnglish
Pages190-202
Number of pages13
JournalRenewable Energy
Volume140
Early online date14 Mar 2019
DOIs
Publication statusPublished - 30 Sep 2019

Fingerprint

Condition monitoring
Wind turbines
Exponential functions
Fault detection
Chemical analysis
Uncertainty

Keywords

  • gaussian process
  • wind turbine
  • SCADA analysis
  • power curve
  • machine learning
  • covariance function

Cite this

@article{469eddd875304fa5a0b3cc45857955bf,
title = "Comparative analysis of Gaussian Process power curve models based on different stationary covariance functions for the purpose of improving model accuracy",
abstract = "Gaussian Process (GP) models are increasingly finding application in wind turbine condition monitoring and in particular early fault detection. GP model accuracy is greatly influenced by the choice and type of the covariance functions (used to described the similarity between two given data points). Hence, the appropriate selection and composition of covariance functions is essential for accurate GP modelling.In this study, an in-depth analysis of commonly used stationary covariance functions is presented in which wind turbine power curve used where GP based power curve has been constructed using different stationary covariance functions, and after that, a comparative analysis has been carried out in order to identify the most effective covariance function. The commonly used squared exponential covariance function is taken as the benchmark, against which other covariance functions are assessed.The results show that the performance (in terms of model accuracy and uncertainty) of GP fitted power curve models based on rational quadratic covariance functions is almost the same as for the most commonly used squared exponential function. Thus, rational quadratic covariance functions can be used instead of squared exponential covariance functions. In this paper, strength and weakness of stationary covariance functions would be highlighted for effective condition monitoring.",
keywords = "gaussian process, wind turbine, SCADA analysis, power curve, machine learning, covariance function",
author = "Pandit, {Ravi Kumar} and David Infield",
year = "2019",
month = "9",
day = "30",
doi = "10.1016/j.renene.2019.03.047",
language = "English",
volume = "140",
pages = "190--202",
journal = "Renewable Energy",
issn = "0960-1481",

}

TY - JOUR

T1 - Comparative analysis of Gaussian Process power curve models based on different stationary covariance functions for the purpose of improving model accuracy

AU - Pandit, Ravi Kumar

AU - Infield, David

PY - 2019/9/30

Y1 - 2019/9/30

N2 - Gaussian Process (GP) models are increasingly finding application in wind turbine condition monitoring and in particular early fault detection. GP model accuracy is greatly influenced by the choice and type of the covariance functions (used to described the similarity between two given data points). Hence, the appropriate selection and composition of covariance functions is essential for accurate GP modelling.In this study, an in-depth analysis of commonly used stationary covariance functions is presented in which wind turbine power curve used where GP based power curve has been constructed using different stationary covariance functions, and after that, a comparative analysis has been carried out in order to identify the most effective covariance function. The commonly used squared exponential covariance function is taken as the benchmark, against which other covariance functions are assessed.The results show that the performance (in terms of model accuracy and uncertainty) of GP fitted power curve models based on rational quadratic covariance functions is almost the same as for the most commonly used squared exponential function. Thus, rational quadratic covariance functions can be used instead of squared exponential covariance functions. In this paper, strength and weakness of stationary covariance functions would be highlighted for effective condition monitoring.

AB - Gaussian Process (GP) models are increasingly finding application in wind turbine condition monitoring and in particular early fault detection. GP model accuracy is greatly influenced by the choice and type of the covariance functions (used to described the similarity between two given data points). Hence, the appropriate selection and composition of covariance functions is essential for accurate GP modelling.In this study, an in-depth analysis of commonly used stationary covariance functions is presented in which wind turbine power curve used where GP based power curve has been constructed using different stationary covariance functions, and after that, a comparative analysis has been carried out in order to identify the most effective covariance function. The commonly used squared exponential covariance function is taken as the benchmark, against which other covariance functions are assessed.The results show that the performance (in terms of model accuracy and uncertainty) of GP fitted power curve models based on rational quadratic covariance functions is almost the same as for the most commonly used squared exponential function. Thus, rational quadratic covariance functions can be used instead of squared exponential covariance functions. In this paper, strength and weakness of stationary covariance functions would be highlighted for effective condition monitoring.

KW - gaussian process

KW - wind turbine

KW - SCADA analysis

KW - power curve

KW - machine learning

KW - covariance function

UR - https://www.sciencedirect.com/journal/renewable-energy

U2 - 10.1016/j.renene.2019.03.047

DO - 10.1016/j.renene.2019.03.047

M3 - Article

VL - 140

SP - 190

EP - 202

JO - Renewable Energy

T2 - Renewable Energy

JF - Renewable Energy

SN - 0960-1481

ER -