Comparison of binned and Gaussian Process based wind turbine power curves for condition monitoring purposes

Research output: Contribution to journalArticle

Abstract

Performance monitoring based on available SCADA data is a cost effective approach to wind turbine condition appraisal. A power curve of a wind turbine describes the relationship between power output and wind speed and is a key measure of wind turbine performance. The standard IEC method calculates a binned power curve from extensive measured data, however this approach requires an extended measurement period in order to limit the uncertainty associated with the calculated power curve, and is far too slow to be used directly for condition monitoring where any changes in operation need to be identified quickly. Hence an efficient approach needs to be developed to overcome this limitation and be able to detect anomalies quickly, thus detecting damage at an early stage so as to prevent catastrophic damage. A Gaussian Process (GP), which is a non-parametric machine learning approach, has the potential fit power curves quickly and effectively. This paper deals with the application of a Gaussian Process to power curve fitting and anomaly detection. This is compared with the conventional approach based on a binned power curve together with individual bin probability distributions to identify operational anomalies. The paper will outline the advantages and limitations of the Gaussian Process approach.

Fingerprint

Condition monitoring
Wind turbines
Curve fitting
Bins
Probability distributions
Learning systems
Monitoring
Costs

Keywords

  • wind turbines
  • Gaussian process
  • condition monitoring
  • power curve
  • curve fitting
  • anomaly detection
  • turbine performance

Cite this

@article{29dc1a54117d4123930fe7aaf789a736,
title = "Comparison of binned and Gaussian Process based wind turbine power curves for condition monitoring purposes",
abstract = "Performance monitoring based on available SCADA data is a cost effective approach to wind turbine condition appraisal. A power curve of a wind turbine describes the relationship between power output and wind speed and is a key measure of wind turbine performance. The standard IEC method calculates a binned power curve from extensive measured data, however this approach requires an extended measurement period in order to limit the uncertainty associated with the calculated power curve, and is far too slow to be used directly for condition monitoring where any changes in operation need to be identified quickly. Hence an efficient approach needs to be developed to overcome this limitation and be able to detect anomalies quickly, thus detecting damage at an early stage so as to prevent catastrophic damage. A Gaussian Process (GP), which is a non-parametric machine learning approach, has the potential fit power curves quickly and effectively. This paper deals with the application of a Gaussian Process to power curve fitting and anomaly detection. This is compared with the conventional approach based on a binned power curve together with individual bin probability distributions to identify operational anomalies. The paper will outline the advantages and limitations of the Gaussian Process approach.",
keywords = "wind turbines, Gaussian process, condition monitoring, power curve, curve fitting, anomaly detection, turbine performance",
author = "Pandit, {Ravi Kumar} and David Infield",
note = "Paper presented at the Second International Conference on Maintenance Engineering (IncoME 2017).",
year = "2018",
month = "9",
day = "1",
language = "English",
volume = "2",
journal = "Journal of Maintenance Engineering",
issn = "1355-2511",

}

TY - JOUR

T1 - Comparison of binned and Gaussian Process based wind turbine power curves for condition monitoring purposes

AU - Pandit, Ravi Kumar

AU - Infield, David

N1 - Paper presented at the Second International Conference on Maintenance Engineering (IncoME 2017).

PY - 2018/9/1

Y1 - 2018/9/1

N2 - Performance monitoring based on available SCADA data is a cost effective approach to wind turbine condition appraisal. A power curve of a wind turbine describes the relationship between power output and wind speed and is a key measure of wind turbine performance. The standard IEC method calculates a binned power curve from extensive measured data, however this approach requires an extended measurement period in order to limit the uncertainty associated with the calculated power curve, and is far too slow to be used directly for condition monitoring where any changes in operation need to be identified quickly. Hence an efficient approach needs to be developed to overcome this limitation and be able to detect anomalies quickly, thus detecting damage at an early stage so as to prevent catastrophic damage. A Gaussian Process (GP), which is a non-parametric machine learning approach, has the potential fit power curves quickly and effectively. This paper deals with the application of a Gaussian Process to power curve fitting and anomaly detection. This is compared with the conventional approach based on a binned power curve together with individual bin probability distributions to identify operational anomalies. The paper will outline the advantages and limitations of the Gaussian Process approach.

AB - Performance monitoring based on available SCADA data is a cost effective approach to wind turbine condition appraisal. A power curve of a wind turbine describes the relationship between power output and wind speed and is a key measure of wind turbine performance. The standard IEC method calculates a binned power curve from extensive measured data, however this approach requires an extended measurement period in order to limit the uncertainty associated with the calculated power curve, and is far too slow to be used directly for condition monitoring where any changes in operation need to be identified quickly. Hence an efficient approach needs to be developed to overcome this limitation and be able to detect anomalies quickly, thus detecting damage at an early stage so as to prevent catastrophic damage. A Gaussian Process (GP), which is a non-parametric machine learning approach, has the potential fit power curves quickly and effectively. This paper deals with the application of a Gaussian Process to power curve fitting and anomaly detection. This is compared with the conventional approach based on a binned power curve together with individual bin probability distributions to identify operational anomalies. The paper will outline the advantages and limitations of the Gaussian Process approach.

KW - wind turbines

KW - Gaussian process

KW - condition monitoring

KW - power curve

KW - curve fitting

KW - anomaly detection

KW - turbine performance

UR - http://www.mace.manchester.ac.uk/our-research/seminars/income-2017/

M3 - Article

VL - 2

JO - Journal of Maintenance Engineering

T2 - Journal of Maintenance Engineering

JF - Journal of Maintenance Engineering

SN - 1355-2511

ER -