LETTER • OPEN ACCESS You may also like

- Nitrogen and carbon limitation of

Using machine learning to predict wind turbine planktonic primary production and
phytoplankton–bacterioplankton coupling
power output in ponds on the McMurdo Ice Shelf,
Antarctica
Brian K Sorrell, Ian Hawes and Karl Safi
To cite this article: A Clifton et al 2013 Environ. Res. Lett. 8 024009 - Why do forest products become less
available?A pan-tropical comparison of
drivers of forest-resource degradation
Kathleen Hermans-Neumann, Katharina
Gerstner, Ilse R Geijzendorffer et al.
View the article online for updates and enhancements. - Compound hot temperature and high
chlorophyll extreme events in global lakes
R Iestyn Woolway, Benjamin M Kraemer,
Jakob Zscheischler et al.

This content was downloaded from IP address 45.186.255.242 on 22/03/2022 at 11:33

IOP PUBLISHING ENVIRONMENTAL RESEARCH LETTERS
Environ. Res. Lett. 8 (2013) 024009 (8pp) doi:10.1088/1748-9326/8/2/024009

Using machine learning to predict wind

turbine power output
A Clifton1 , L Kilcher1 , J K Lundquist1,2 and P Fleming1
1
National Wind Technology Center, National Renewable Energy Laboratory, Golden, CO 80401, USA
2
Department of Atmospheric and Oceanic Sciences, University of Colorado at Boulder, Boulder,
CO 80309, USA

E-mail: andrew.clifton@nrel.gov

Received 22 December 2012

Accepted for publication 5 March 2013
Published 18 April 2013
Online at stacks.iop.org/ERL/8/024009
Abstract
Wind turbine power output is known to be a strong function of wind speed, but is also affected
by turbulence and shear. In this work, new aerostructural simulations of a generic 1.5 MW
turbine are used to rank atmospheric influences on power output. Most significant is the hub
height wind speed, followed by hub height turbulence intensity and then wind speed shear
across the rotor disk. These simulation data are used to train regression trees that predict the
turbine response for any combination of wind speed, turbulence intensity, and wind shear that
might be expected at a turbine site. For a randomly selected atmospheric condition, the
accuracy of the regression tree power predictions is three times higher than that from the
traditional power curve methodology. The regression tree method can also be applied to
turbine test data and used to predict turbine performance at a new site. No new data are
required in comparison to the data that are usually collected for a wind resource assessment.
Implementing the method requires turbine manufacturers to create a turbine regression tree
model from test site data. Such an approach could significantly reduce bias in power
predictions that arise because of the different turbulence and shear at the new site, compared to
the test site.

Keywords: machine learning, classification and regression trees, wind energy, wind turbine

1. Introduction Commission 2011). As a result, it is essential to accurately

forecast the output power from wind turbines both before they
Wind turbines are an important part of the electricity are installed and once they are operational.
generation portfolio worldwide. In 2011, wind turbines A wind turbine’s blades sweep through a circular disk,
known as the rotor disk (figure 1). The power output by a
provided 3% of the power generation capacity in the United
wind turbine is a function of the kinetic energy flux through
States, and six US states in the USA met more than 10% of
the rotor disk and the efficiency with which that energy can
their annual electricity demand through wind energy (Wiser
be captured. If the wind has an instantaneous speed u that
and Bollinger 2012). In that year, the rate of installation of is uniform throughout a rotor disk of diameter d, the power
new wind turbine generation capacity was second only to captured is:
natural gas. Some markets have higher penetration of wind
energy; for example, wind turbines provided 3.5% of the 1 d2
PK = cp ρπ u3 , (1)
energy generated in the European Union in 2008 (European 2 4
where ρ is the air density and cp is the ‘power coefficient’,
Content from this work may be used under the terms of
the Creative Commons Attribution 3.0 licence. Any further
which describes the fraction of energy that a wind turbine
distribution of this work must maintain attribution to the author(s) and the captures. The best theoretical performance of a turbine is the
title of the work, journal citation and DOI. Betz limit, at which cp = 16/27.

1748-9326/13/024009+08$33.00 1 c 2013 IOP Publishing Ltd Printed in the UK

Environ. Res. Lett. 8 (2013) 024009 A Clifton et al

Figure 2. Power curves for several different utility-scale wind

turbines. Data are from the manufacturers’ power curve tables,
collated by Idaho National Laboratory (Seifert 2007).

Figure 1. Schematic of wind turbine performance test. Cup

anemometers mounted on a mast two to four rotor diameters upwind potentially impact the power captured (1). Previous studies
of the turbine measure wind speed at several heights across the rotor have shown that variation in atmospheric conditions can
disk, including hub height. lead to changes in turbine power output of 10% or more at
the same wind speed (Antoniou et al 2009, Wagner et al
2011, Hedevang 2012, Vanderwende and Lundquist 2012,
However, real wind turbines do not achieve this Wharton and Lundquist 2012). However, only a few turbine
theoretical limit. Their performance is a function of manufacturers provide different power curves for different Ti,
aerodynamics and the need to limit power capture once suggesting that the effect of turbulence and shear on power
the rated generator power is reached, at ‘rated’ wind output is a recognized effect that is difficult to include in
speed. The generator power, turbine diameter and blade turbine power predictions.
shape are optimized based on site characteristics such as The effects of changes in flow on power capture can
annual average wind speed and the wind speed distribution. be significant. Because of intermittency in the wind, wind
Turbine manufacturers measure their turbine’s ‘power turbines typically produce 20%–40% of their maximum
curve’—the relationship between power output and wind possible output over the course of a year. This is known
speed—at turbine test sites (figure 1). The power curve is as the capacity factor of the turbine and varies by turbine
calculated from 10 min averaged wind speed (U = ū) and and site (Manwell et al 2009). If a wind turbine has a
power using methods described in standards (International baseplate capacity of 1.5 MW and a capacity factor of 20%,
Electrotechnical Commission 2005). Typical power curves it will generate 2628 MWh of energy per year. If the energy
have an s-shape (figure 2), where at wind speeds less than generated is sold at $30–$60 per MWh (Wiser and Bollinger
rated the energy capture is approximately proportional to U 3 2012), the loss of just 1% from this turbine compared
(known as Region II). At wind speeds above rated, the blade to predicted power generation represents $788–$1577 lost
pitch and generator torque are actively controlled to limit the income per year, per turbine. This uncertainty may be
power to the generator’s rated power (Region III). particularly important when predicting the power at a
Other metrics provide more information about the inflow potential turbine installation site using local wind speed data.
characteristics at the test site. Turbulence is quantified by If the amount of energy that is available is over-estimated,
turbulence intensity Ti = σ (u(t))/U, where σ (u(t)) is the the resulting calculation of the cost of electricity may be too
standard deviation of the wind speed during a 10 min interval. optimistic and the site may not be as profitable as expected.
Variation in wind speed with height z can be quantified by Alternatively, underestimating the energy available at a site
fitting the power-law profile U(z) = βzα to measurements. may mean that the site is not developed at all.
The power-law exponent α is used in this study and the wind This paper describes how data that are typically collected
industry to quantify wind shear (Brower 2012). during testing of a wind turbine may be used to create a more
The inverse of the turbine testing process is applied to accurate and robust model of the wind turbine energy capture,
predict how the same wind turbines will perform at a new site; which can then be used at a new site. Section 2 shows how
the wind speed is measured at the turbine hub height and then the energy captured by a generic wind turbine changes with
used to predict power from the power curve. Ti and α are only different inflow conditions. In section 3, a machine learning
checked to confirm that the new site is within the limits for the algorithm is introduced that can be used to more accurately
turbine tests. predict the energy capture of the wind turbine under differing
Any change in the atmosphere that results in a change site conditions. Section 4 discusses potential applications for
of the kinetic energy through the turbine rotor disk will this technique.

2
Environ. Res. Lett. 8 (2013) 024009 A Clifton et al

Table 1. Design characteristics of the WindPACT 1.5 MW baseline

wind turbine. Sources: Poore and Lettenmaier (2003) and Malcolm
and Hansen (2006).
Parameter Value
Hub height 84 m
Rotor diameter 70 m
Cut-in speed 3 m s−1
Rated speed 11.5 m s−1
Cut-out speed 27.6 m s−1
Rated power 1.5 MW

2. Response of a wind turbine to turbulence and

shear
Figure 3. Power curve and operating regions for the WindPACT
Wind turbines with a capacity from 1 to 2 MW represented 1.5 MW turbine. Data are derived from 1796 FAST simulations
58% of the total number of turbines over 100 kW that forced by TurbSim wind fields. Power data are binned into 1 m s−1
wind speed bins. The reference zero-turbulence case is plotted
were installed in the United States between 2002 and 2011 together with the data from the turbulent flow fields. The median
(Wiser and Bollinger 2012). Therefore, we use simulations power produced in the turbulent flow is marked with a horizontal
of a generic 1.5 MW wind turbine to show how changes blue line. The box covers the interquartile range. Whiskers extend to
in inflow conditions may impact turbine performance. The the 5th and 95th percentiles, and individual outliers are marked.
turbine design we use is from the Wind Partnerships for
Advanced Component Technology (WindPACT) study and
is broadly representative of variable speed, variable pitch, Data from the database of simulations can be used
1.5 MW turbines (table 1). to create a power curve by binning the power data into
1 m s−1 -wide bins (figure 3). Although the forcing variables
This study uses the FAST aerostructural simulator
are evenly distributed, variance in power is largest near rated
(Jonkman and Buhl 2005) to simulate the aerodynamic forces
wind speed. This sensitivity may result in large variation
on the turbine blades and structure. FAST uses blade-element
between predicted power output and observed power output.
momentum theory, which is a physically based method that is
Furthermore, the mean power generated in simulations that
widely used in the wind industry to create quick and accurate
include turbulence is lower than the no-turbulence cases near
simulations of turbine performance. The aerodynamic lift and
rated wind speed. The effects of turbulence observed in
drag are calculated at different points along the blade from
simulations agree with field studies (e.g. Vanderwende and
the incoming wind speed and the blade profile together with
Lundquist 2012). These results suggest that power predictions
look-up tables or simple aerodynamic solvers. As with a real
at a new site might be biased if the wind conditions (Ti, α and
turbine, forces acting on the blade cause rotation of the shaft their co-dependency with wind speed U) are different than the
and deflection of the structure. The rotation is converted into test site where the power curve was measured. For example,
electrical power in a generator. Results from FAST have been mountain sites might have quite different characteristics than
validated against data from several wind turbine designs. coastal or inshore sites (e.g. Sathe et al 2012, Clifton and
For this study, a total of 1796 10 min wind fields were Lundquist 2012).
created using the stochastic turbulence simulator, TurbSim Power output normalized by the zero-turbulence power
(Jonkman 2009, Kelley 2011). TurbSim creates wind fields curve (‘normalized power’) shows a clear dependence on
that include wind shear and turbulence, where the power shear exponent and turbulence intensity (figure 4). At wind
spectrum of the turbulence approximates the Kaimal spectrum speeds below 8 m s−1 , power increases with turbulence
assumed in wind turbine design standards. TurbSim can take intensity and shear (figure 4(a)). The increase in power due
target hub height values or profiles of U, α, and Ti as inputs. to turbulence arises because turbulent flow with mean speed
In this study, wind fields have random combinations of hub U carries more power than laminar flow of the same U, and
height wind speed from 3 to 25 m s−1 , hub height turbulence is well documented in literature (e.g. Wagner et al 2010). The
intensity from 5% to 45%, and shear exponent from −0.5 changes in power output of ±20% associated with turbulence
to 0.5. All variables are uniformly randomly distributed over are approximately half of the change due to a change in wind
these ranges. These wind fields were then used to force the speed from 7 to 8 m s−1 (1).
FAST model of the WindPACT 1.5 MW baseline turbine. A In contrast, at wind speeds just above and below rated
reference set of zero-turbulence simulations with Ti = 0% and speed (figures 3, 4(b) and (c)), increasing turbulence intensity
α = 0.2 was also created. reduces power output as the turbine cannot capture the extra
The data from the wind fields from TurbSim and the energy that gusts bring, but a short duration slow down to
output from FAST were then processed to form a database wind speeds below rated results in a loss of energy. As the
of 1796 observations of 10 min average power (the response) mean wind speed increases, the total amount of time with the
as a function of wind speed, turbulence intensity, and shear blades pitched toward feather increases and the wind turbine
exponent (the forcing). is more often operating at rated power. At wind speeds much

3
Environ. Res. Lett. 8 (2013) 024009 A Clifton et al

Figure 5. Visualization of the process of predicting the 10 minute

mean power for a single observation of U, Ti, α and region. The
visualization shows one branch of one member of a regression tree.
The predictions in this study use an ensemble of 100 regression
trees.

combinations of these variables, a turbine test site might not

capture all possible combinations that might be found at a new
site. Therefore, it is desirable that a method can interpolate
between test observations.
Figure 4. Simulation-derived power output of the WindPACT The power output from the turbine is not a linear function
1.5 MW wind turbine under different inflow conditions, normalized of wind speed (figure 3), so, multivariate linear regression is
by the zero-turbulence case. Colors show the median power not an appropriate technique. Non-linear regression assumes
produced in bins of Ti = 2% by α = 0.05, using the color scale on that the relationships are constant throughout the model space
the right. (a) 7 ≤ U < 8 m s−1 . (b) 12 ≤ U < 13 m s−1 .
(c) 16 ≤ U < 17 m s−1 . (i.e. power output is always proportional to U n ), which
from figure 3 is incorrect, so non-linear regression is also
inappropriate. Also, multivariate bins only work where the
greater than rated (figure 4(c)), larger turbulence intensities training data includes data in all bins and so would be
are required to reduce the output of the machine to less computationally or observationally more expensive. Instead,
than rated power, regardless of shear. In Regions II and III, a technique is required that can capture non-linear changes in
variation in Ti impacts power performance more than variation response to forcing. For this reason, we propose a machine
in α, as also seen in Wharton and Lundquist (2012). learning technique called ‘regression trees’ (Breiman et al
1984) that is implemented in the MATLAB Statistics Toolbox
3. Using machine learning to predict power output (MATLAB 2010). Regression trees are models that use simple
in different conditions branching question paths to predict an outcome based on
inputs. The precise arrangement of the branches depends on
3.1. Model design the data that are used to train the model. A sketch of one
branch of one tree is shown in figure 5, where a prediction
Currently, we seek a robust method for incorporating of power output is made for a set of forcing conditions.
turbulence intensity and wind shear into power prediction This study uses the mean value predicted by an ensemble
tools. Just as the database of forcing conditions (hub height of 100 regression trees to increase the accuracy of the
wind speed, turbulence intensity, and wind shear) and power predictions. In the ensemble method, subsets of the
turbine response (power output) does not contain all possible training data are used to generate different trees, and power

4
Environ. Res. Lett. 8 (2013) 024009 A Clifton et al

is calculated using each tree, forced with the same input. The
outputs from the 100 different trees are combined to give the
mean and variance of the power estimate. An ensemble of
trees reduces variance and is less affected by ‘noisy’ training
data, than a single tree.
Hub height wind speed U, hub height turbulence intensity
Ti, wind shear exponent α, and operating region are chosen
as the predictive variables, while the 10 min mean power is
the predicted value. All of these variables can be sampled
during turbine testing. Importantly, the predictive variables
(U, Ti, α) may be available for a new site if high-frequency
hub height wind data are collected at the new location. The
turbine operating region can be estimated as a function of
mean wind speed and turbine rated speed. Other variables may
be important for turbine performance but are not considered in
this first exploration of the regression tree method.

3.2. Interpolating between observations

The regression tree model can be used to create a similar

visualization of the wind turbine performance that was shown
in figure 4. Figure 6 shows the modeled effect of shear and
turbulence on the 10 min average power at three different hub
height wind speeds. Compared to the data shown in figure 4,
figure 6 shows clear gradients in power output as a function of
turbulence intensity or shear. The authors’ experience is that
predicting power output for a specific combination of inflow
conditions using a regression tree requires approximately
one-millionth the CPU hours of a simulation using TurbSim
and FAST.
The WindPACT turbine responds differently to changes
in shear and turbulence at different wind speeds. In Region
II, at wind speeds below 8 m s−1 (figure 6(a)), power output
increases by up to 10% as turbulence increases or as the
magnitude of the shear increases. At wind speeds greater than
8 m s−1 and in Region III (figures 6(b) and (c)) the regression
tree modeled power is consistent with the simulated power Figure 6. Model-derived power output of the WindPACT 1.5 MW
output: power decreases as Ti increases and shows weak or no wind turbine under different inflow conditions. Data are shown for a
single wind speed in each subfigure. Colors show the median power
dependence on shear. output in bins of Ti = 2% by α = 0.05, according to the color scale
The larger change in power for |α| > 0.45 seen in figure 6 on the right. Contour lines show the power normalized by the
suggests a poor performance of the regression tree approach zero-turbulence case with α = 0.2. The interval between contour
at the edge of the data range. For this reason, the regression lines is 2.5%. (a) U = 7.5 m s−1 . (b) U = 12.5 m s−1 .
tree model should probably only be used where |α| < 0.45. (c) U = 16.5 m s−1 .
The limit on the range of α is due to the range of the training
data and is not inherent to the regression tree approach, and
shows that users should be careful to check that the test site 0.0 < α < 0.25. This overlap between training and site data is
data captures the range of variability that might be seen at not always possible, and because there may be other factors
potential deployment sites. that affect energy capture, there is some uncertainty in the
forecast of power capture. To quantify the accuracy of the
3.3. Predicting turbine response to unseen conditions energy capture forecast, the 1796 simulations are randomly
divided into a 898-member training dataset and a 898-member
The regression tree model can predict the power output by validation dataset. This random selection is repeated 50 times
the wind turbine in previously unseen conditions. Predicting to give 50 individual test cases. In each test case, the training
energy capture in new conditions ideally requires that the data data is used to create a power curve and to train a regression
that were used to train the model encompass the new set of tree model to predict the 10 min mean power. The power
conditions. For example, if the new site has 5 < Ti < 10%, curve is created by calculating the bin-mean power output in
and 0.1 < α < 0.2, it would be ideal if the training data 1 m s−1 -wide bins from 3 to 4 m s−1 , 4–5 m s−1 , etc. These
set included data that have the range 3 < Ti < 12% and models are used to predict the power output (pi ) for each of

5
Environ. Res. Lett. 8 (2013) 024009 A Clifton et al

that the response of the turbine is a complex, non-linear

function of wind speed, turbulence intensity, and shear. The
effect of variations in shear and turbulence intensity is larger
below rated speed than at rated wind speeds. At all wind
speeds, turbulence has more of an effect on performance
than shear. If the WindPACT 1.5 MW behavior is typical of
modern turbines, there can be 5–10% variation in 10 min
mean power at the same hub height wind speed in typical
atmospheric conditions of 5 < Ti < 20% and −0.1 < α <
0.3. As conditions vary further, the power output deviates
even more. Other atmospheric effects such as directional veer,
turbulence kinetic energy, and the presence of a low-level jet
and associated coherent structures (e.g. Kelley 2011) may also
impact the power output of the wind turbine. The condition of
the wind turbine itself may also be important; for example
errors in the orientation of the turbine with respect to the wind
(‘yaw error’, e.g. Kragh et al 2013), or changes in the turbine
control system. The regression tree method could be extended
to include other atmospheric effects and more information
about the state of the wind turbine if that data were available
Figure 7. Root-mean-square error (RMSE) and mean absolute
error (MAE) for predictions of 10 minute mean power for 50 test from more complex simulations, or using data from a turbine
cases using the power curve or regression tree approaches. Test test sites such as the National Wind Technology Center in
cases use training and validation data sets containing half of the Colorado (Clifton et al 2012) or in Europe (e.g. Sathe et al
1796 simulations. Data are shown versus the mean power from the 2012).
simulations in each test case. This study shows that the accuracy of the current power
curve method may depend on the distribution of wind speed,
the n = 898 members of the validation data set. The power turbulence intensity, and shear at the test site, compared to
output of each member is known from the simulations (Pi ). the deployment site. If the test site conditions are similar
Metrics used to quantify the accuracy of the predicted power to the deployment site, the power curve method may give
from the power curve and regression tree model in each test good results. The greatest potential for error when using a
case include the root-mean-squared error (RMSE) and mean power curve approach comes when the test site has high hub
absolute error (MAE): height turbulence and high shear compared to the deployment
site (figures 4 and 6). In this study, the simulation data have
v
u n
u1 X  been used to train a regression tree model of the WindPACT
(pi − Pi )2 ,

RMSE = t (2) 1.5 MW baseline turbine. The regression tree method predicts
n i=1 wind turbine energy capture with two to three times more
n accuracy than the industry-standard power curve method, and
1X may be more useful for predictions of energy capture at sites
MAE = |pi − Pi | . (3)
n i=1 that experience different conditions than the test site.
To use the regression tree modeling approach to predict
RMSE and MAE are commonly used to judge forecast
the energy capture of a turbine at a new site, several steps
accuracy. Lower RMSE or MAE indicates better prediction of
are required. First, a regression tree model of a particular
the turbine response.
wind turbine is created from test data by the manufacturer
Figure 7 summarizes the error metrics for the 50 test cases
or a testing agency. The test data would include inflow
versus the mean power in each test case. The RMSE and MAE data (e.g. U, Ti, and α) and turbine power data, which
of the power prediction using the power curve derived from are all collected during tests that conform to international
binned power data was typically two to three times higher than standards (International Electrotechnical Commission 2005).
using the regression tree method. Although the power curve Then, the same inflow data is measured at the wind turbine
method achieved R2 > 0.99 between predicted and observed deployment site as part of the wind resource assessment
power, the regression tree method achieved R2 > 0.995 in all process. Usually only one year of data is collected at
cases (not shown). the deployment site and extrapolated by comparison to a
local long-term measurement site to give a prediction of
4. Discussion the wind speed for 25 years (Brower 2012). This process
of extrapolation is called measure–correlate–predict and is
Aerostructural simulations of a generic 1.5 MW wind well-defined for wind speed, but there is no well-known
turbine in a turbulent flow field have been used to quantify method for the extrapolation of shear and turbulence intensity.
that turbine’s power output in a range of wind speeds, Instead, it may be necessary to create an artificial 25-year time
turbulence intensities and shear. These simulations show series of forcing conditions from the site observations (see e.g.

6
Environ. Res. Lett. 8 (2013) 024009 A Clifton et al

Rose and Apt 2012). Finally, the regression model is forced Acknowledgments
with the inflow conditions at a new site to estimate the power
production over the 25 year period. This work was supported by the US Department of Energy
As well as predicting the response of a single wind under Contract No. DE-AC36-08GO28308 with the National
turbine, regression trees could be used to predict energy Renewable Energy Laboratory. Marshall Buhl provided
produced by multiple turbines at a wind plant. The regression valuable support with the initial preparation of FAST
tree could be trained using observations of inflow conditions, simulations and interpretation of FAST data.
turbine availability, and power produced by individual
turbines. The trained model could then be used to predict References
power output either for the individual turbines, or for the
whole site. Where site data does not include Ti and α, proxies Antoniou I, Pedersen S M and Enevoldsen P B 2009 Wind shear
and uncertainties in power curve measurement and wind
such as time of day, cloud cover or wind direction could be resources Wind Eng. 33 449–68
used instead. As with other machine-learning tools applied for Barbounis T, Theocharis J, Alexiadis M and Dokopoulos P 2006
meteorological purposes such as Kalman Filters (e.g. Delle Long-term wind speed and power forecasting using local
Monache et al 2011) or artificial neural networks (Barbounis recurrent neural network models IEEE Trans. Energy Convers.
21 273–84
et al 2006), the use of a regression tree requires only that the Breiman L, Friedman J, Olshen R and Stone C 1984 Classification
inputs to the tree capture most of the variability of the process, and Regression Trees (Boca Raton, FL: CRC Press)
rather than requiring a model that describes the physics. As Brower M 2012 Wind Resource Assessment: A Practical Guide to
a result, they are quick to run and easy to update with new Developing a Wind Project (New York: Wiley)
information, thus, well-suited to forecasting applications. Clifton A and Lundquist J K 2012 Data clustering reveals climate
impacts on local wind phenomena J. Appl. Meteorol. Climatol.
51 1547–57
5. Conclusions Clifton A, Schreck S, Scott G, Kelley N D and Lundquist J 2012
Turbine inflow characterization at the National Wind
Technology Center J. Sol. Energy Eng. accepted
Simulations of a utility-scale wind turbine have been used Delle Monache L, Nipen T, Liu Y, Roux G and Stull R 2011
to develop a database and to show the complex response Kalman filter and analog schemes to postprocess numerical
of the wind turbine to changes in hub height wind speed, weather predictions Monthly Weather Rev. 139 3554–70
turbulence intensity and rotor disk shear. The database of European Commission 2011 Renewables make the difference Tech.
Rep. European Commission
forcing conditions has been combined with the turbine rated Hedevang E 2012 Wind turbine power curves incorporating
speed to define the turbine operating region. Together, these turbulence intensity Wind Energy at press (doi:10.1002/we.
data were used to generate a regression-tree model of the 1566)
wind turbine’s power generation. The simulations suggest International Electrotechnical Commission 2005 IEC 61400-12:
and the model clearly demonstrates that the response of the Wind Turbines—Part 12: Power Performance Measurements of
Electricity Producing Wind Turbines 1st PPUB edn (Geneva,
turbine is a complex non-linear function of hub height wind Switzerland: International Electrotechnical Commission) p 12
speed, turbulence intensity, and rotor disk shear. At wind Jonkman B J 2009 TurbSim user’s guide: version 1.50 Tech. Rep.
speeds below rated speed, the turbine power output is most TP-500-46198 (Golden, CO: National Renewable Energy
sensitive to changes in wind speed and speed, turbulence. Laboratory)
Jonkman J and Buhl M J 2005 FAST user’s guide Tech. Rep.
At rated speed, the turbine is most sensitive to turbulence EL-500-29798 (Golden, CO: National Renewable Energy
intensity and shear, and power can change by 10% under Laboratory)
typical atmospheric conditions. At wind speeds greater than Kelley N D 2011 Turbulence–turbine interaction: the basis for the
rated, the turbine responds most to changes in turbulence development of the TurbSim stochastic simulator Tech. Rep.
intensity. TP-5000-52353 (Golden, CO: National Renewable Energy
Laboratory)
Predictions of power output using the regression tree Kragh K A, Fleming P A and Scholbrock A 2013 Increased power
model are approximately three times more accurate than capture by rotor speed dependent yaw control of wind turbines
power predictions using the power curve. Although this J. Sol. Energy Eng. at press
method has been demonstrated using simulated inflow and Malcolm D J and Hansen A 2006 WindPACT turbine rotor design
study Subcontract Report SR-500-32495 (Golden, CO:
turbine response data, the method could be used to generate National Renewable Energy Laboratory)
turbine performance models from turbine power testing data. Manwell J, McGowan J and Rogers A 2009 Wind Energy Explained
Changes of wind direction with height, non-uniform shear, 2nd edn (New York: Wiley)
and the state of the turbine were not considered here but MATLAB 2010 Version 7.11.0 (R2010b) (Natick, MA: The
MathWorks Inc)
may impact turbine deployment sites, and their effect should
Poore R and Lettenmaier T 2003 Alternative design study report:
be investigated using field data. Application of the data to WindPACT advanced wind turbine drive train designs study
wind turbine deployment sites does not require any new Subcontractor Report NREL/SR-500-33196 (Golden, CO:
instrumentation compared to what is currently used. Using National Renewable Energy Laboratory)
this regression tree method helps eliminate biases resulting Rose S and Apt J 2012 Generating wind time series as a hybrid of
measured and simulated data Wind Energy 15 699–715
from deployment site wind conditions (turbulence intensity, Sathe A, Mann J, Barlas T, Bierbooms W and van Bussel G 2012
shear, and their correlations with wind speed) that differ from Influence of atmospheric stability on wind turbine loads Wind
the turbine test site. Energy at press (doi:10.1002/we.1528)

7
Environ. Res. Lett. 8 (2013) 024009 A Clifton et al

Seifert G 2007 Idaho Wind Data (Idaho Falls, ID: Idaho National Wagner R, Courtney M, Larsen T and Paulsen U 2010 Simulation of
Laboratory) shear and turbulence impact on wind turbine performance
Vanderwende B J and Lundquist J K 2012 The modification of wind Tech. Rep. Risø-R-1722(EN) (Roskilde: Risø National
turbine performance by statistically distinct atmospheric Laboratory)
regimes Environ. Res. Lett. 7 034035 Wharton S and Lundquist J K 2012 Atmospheric stability affects
Wagner R, Courtney M, Gottschall J and Lindelöw-Marsden P 2011 wind turbine power collection Environ. Res. Lett. 7 014005
Accounting for the speed shear in wind turbine power Wiser R and Bollinger M 2012 2011 Wind technologies market
performance measurement Wind Energy 14 993–1004 report Tech. Rep. (Washington, DC: US Department of Energy)