Probabilistic modelling of wind turbine power curves with application of heteroscedastic Gaussian Process regression

Rogers, Timothy J.; Gardner, Paul A.; Dervilis, Nikolaos; Worden, Keith; Maguire, Eoghan; Papatheou, Evangelos; Cross, Elizabeth J.

doi:10.1016/j.renene.2019.09.145

Cited by 70 publications

(63 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…GPs have been studied as a regression technique since the early 90s. Initially discovered as being a limiting case of some types of neural networks, as the number of hidden layers becomes infinite, 9 they have since been shown to be a flexible and robust regression technique and have been applied to many different tasks, 10 including in the wind energy context 11,12 …”

Section: A Gp Regression Approachmentioning

confidence: 99%

Wind site turbulence de‐trending using statistical moments: Evaluating existing methods and introducing a Gaussian process regression approach

et al. 2021

View full text Add to dashboard Cite

This paper considers the problem of retrospectively de‐trending wind site data when only statistical moments, in the form of 10‐min means and standard deviations in wind speed, are available. Low‐frequency trends present in wind speed data are known to bias fatigue damage estimates, and, hence, removal of their influence is important for accurate fatigue life estimation. When raw data is available, this procedure is straightforward; however, for many sites, significant quantities of data are available, which contain only statistical moments. Additional value is therefore unlocked if de‐trending can also take place in this context. Existing methods, Models 1 and 2, are introduced, and their performance and viability appraised, respectively. A Gaussian process (GP) regression implementation is also developed, which seeks to incorporate characteristics of real trends extracted from raw data into the fitting procedure via an appropriately chosen lengthscale hyperparameter. Results indicate that Model 2, the recommended method in previous work, suffers from fundamental issues, with the implication that it should no longer be used. Model 1 and GP results are shown to be very similar at the turbulence distribution level. This finding is interpreted as a validation of Model 1 and an indication that it may already be performing as well as can be hoped for, given the information available in the current problem formulation. Theoretical overheads associated with GPs, in addition to the performance similarities mentioned above, lead to Model 1 being recommended as the best approach to moment‐based turbulence de‐trending at this time.

show abstract

Section: A Gp Regression Approachmentioning

confidence: 99%

Wind site turbulence de‐trending using statistical moments: Evaluating existing methods and introducing a Gaussian process regression approach

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Applying this method does not consider the non-linear power to wind relationship (IEC 61400-12-2:2013, IEC 2013. There are a variety of probabilistic methods (Pedersen and Fossen 2012;Rogers et al 2020;Saravanakumar and Jena 2013) and non-parametric (Pedersen and Fossen 2012) that encapsulate the power curve of a wind turbine. This aspect is crucial to effective power modelling.…”

Section: Performance Monitoringmentioning

confidence: 99%

Condition monitoring systems: a systematic literature review on machine-learning methods improving offshore-wind turbine operational management

Black

Richmond

Kolios

2021

International Journal of Sustainable Energy

View full text Add to dashboard Cite

Information is key. Offshore wind farms are installed with supervisory control and data acquisition systems (SCADA) gathering valuable information. Determining the precise condition of an asset is essential on achieving the expected operational lifetime and efficiency. Equipment fault detection is necessary to achieve this. This paper presents a systematic literature review of machine learning methods applied to condition monitoring systems, using both vibration information and SCADA data together. Starting with conventional methods using vibration models, such as Fast-Fourier transforms to five prominent supervised learning regression models; Artificial neural network, support vector regression, Bayesian network, random forest and K-nearest neighbour. This review specifically looks at how conventional vibration data can be combined with SCADA data to determine the assets condition.

show abstract

“…Due to the superior nonlinear fitting ability of many artificial intelligence methods, they have been widely utilized in wind turbine power curve modeling, namely estimating the unknown nonlinear function f (•). Some popular ones are SVR [20], GP [21,22], and ANNs [19,26].…”

Section: Nonparametric Modelsmentioning

confidence: 99%

“…In [20,21], support vector regression (SVR) and Gaussian process (GP) were employed to estimate the real WTPCs. Moreover, GP was also used to obtain probabilistic WTPC in [21,22]. In [23], Wang et al proposed two Bayesian-based models, heteroscedastic and robust spline regression models, to fit deterministic and probabilistic power curves in different seasons, respectively.…”

Section: Introductionmentioning

confidence: 99%

Wind Turbine Power Curve Modeling with a Hybrid Machine Learning Technique

Pei

2019

Applied Sciences

View full text Add to dashboard Cite

A power curve of a wind turbine describes the nonlinear relationship between wind speed and the corresponding power output. It shows the generation performance of a wind turbine. It plays vital roles in wind power forecasting, wind energy potential estimation, wind turbine selection, and wind turbine condition monitoring. In this paper, a hybrid power curve modeling technique is proposed. First, fuzzy c-means clustering is employed to detect and remove outliers from the original wind data. Then, different extreme learning machines are trained with the processed data. The corresponding wind power forecasts can also be obtained with the trained models. Finally, support vector regression is used to take advantage of different forecasts from different models. The results show that (1) five-parameter logistic function is superior to the others among the parametric models; (2) generally, nonparametric power curve models perform better than parametric models; (3) the proposed hybrid model can generate more accurate power output estimations than the other compared models, thus resulting in better wind turbine power curves. Overall, the proposed hybrid strategy can also be applied in power curve modeling, and is an effective tool to get better wind turbine power curves, even when the collected wind data is corrupted by outliers.

show abstract

Probabilistic modelling of wind turbine power curves with application of heteroscedastic Gaussian Process regression

Cited by 70 publications

References 23 publications

Wind site turbulence de‐trending using statistical moments: Evaluating existing methods and introducing a Gaussian process regression approach

Wind site turbulence de‐trending using statistical moments: Evaluating existing methods and introducing a Gaussian process regression approach

Condition monitoring systems: a systematic literature review on machine-learning methods improving offshore-wind turbine operational management

Wind Turbine Power Curve Modeling with a Hybrid Machine Learning Technique

Contact Info

Product

Resources

About