Journal of Physics: Conference

Series

PAPER • OPEN ACCESS You may also like

- Meeting report
Engineering models for turbine wake velocity - Workshop on Intakes of Radionuclides:
deficit and wake deflection. A new proposed Occupational and Public Exposure,
Avignon, 15-18 September 1997
G Etherington, A W Phipps, J D Harrison
approach for onshore and offshore applications et al.

- SRP Meeting: Developments in

To cite this article: Renzo Ruisi and Ervin Bossanyi 2019 J. Phys.: Conf. Ser. 1222 012004 Operational Health Physics, University of
Birmingham, 25-26 March 1998

View the article online for updates and enhancements.

This content was downloaded from IP address 31.184.48.216 on 07/05/2024 at 14:27

WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

Engineering models for turbine wake velocity deficit and wake

deflection. A new proposed approach for onshore and offshore
applications.
Renzo Ruisi, Ervin Bossanyi

DNV GL, One Linear Park, Avon Street, Bristol, BS2 0PS, UK
[email protected], [email protected]

Abstract. An engineering wake model based on the Ainslie model is proposed. The eddy
viscosity term associated with momentum diffusivity is modified to take into account the effects
of atmospheric stability. The parameters used are typically available from high-quality on-site
measurement campaigns and the effects of atmospheric stability are based on empirical models
for the estimation of the Monin-Obukhov length. The dependence on physical quantities only is
particularly advantageous for fast wake modelling, since no further parametrical tuning is needed
for each specific case. The proposed wake model is initially compared to wind tunnel data and
CFD simulations to test the chosen Obukhov lengths for specific flow conditions. Wind farm
production data and concurrent meteorological data at one onshore site are then used to validate
the model for specific on-site flow conditions, obtaining good results. Two offshore wind farms
are also used to assess the model in a large-scale wind farm scenario: results look promising
although some reservations are expressed on the effect of the wake superposition model. Models
for the prediction of wake centreline deflection due to yaw are compared at different yaw angles
using wind tunnels data and CFD simulations. Although the EPFL model showed some
advantage, especially in non-neutral conditions, both models give satisfactory results.
Furthermore, it was showed that the wake deflection caused by the rotating wake for non-yawed
turbines can have a large impact on the predictions of the centreline deflection.

1. Introduction
The need for fast and increasingly accurate wake models for wind turbines is promoted by the quest for
higher accuracy in energy yield predictions, the increased size of wind farms, and by the need to exploit
the potential for increasing wind farm energy production through active wake control methods which
are currently being developed. The increased interest of the industry towards wind farm optimisation
control algorithms, for example wake steering, highlights even more the importance of reliable wake
models which can run fast enough for control design optimisation and testing.
Numerous efforts have resulted in various wake models available in literature: from the seminal
works of Lissaman [1], Jensen [2] and Ainslie [3, 4], to the more recent works of Bastankhah & Porté-
Agel [5] and Niayfair & Porté-Agel [6], Ishihara [7] and Gebraad [8, 9, 10]. These wake models have
been developed and validated against wind farm production data, computational simulations and wind
tunnel data. . Engineering wake models rely, to different extents, on wake measurements and subsequent
fitting of empirical parameters, which exposes the models to the risk of over-fitting, potentially
producing erroneous results when extrapolated to different configurations of turbine characteristics,
wind farm layouts and atmospheric conditions.
More advanced models based on a Computational Fluid Dynamics (CFD) approach have been used
in the recent years. From simulations based on Reynolds-Averaged-Navier-Stokes (RANS) equations
and related turbulence models, to the more complex use of Large-Eddy-Simulations (LES) using
Actuator-Line turbine models [11, 12, 13]. Their main common drawback is the need of large
computational resources and time, which does not suit the market needs for a reliable prediction of wake
losses during the planning phase of a wind farm, usually including different turbine models and
configurations.

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

In the wind energy industry, it is common to use RANS simulations associated to the 𝑘 − 𝜖
turbulence model and generally using an Actuator Disk Model (abbreviated as ADM-R or ADM-NR,
where the suffix R indicates a rotating disk model and NR indicates a non-rotating disk model) to allow
the calculation of loads and wind speed deficits due to the presence of the rotor. It is also noted that
amendments to the 𝑘 − 𝜖 model have been recently proposed for the use in wind turbine applications in
[13]. A more computationally expensive approach, used for high-fidelity analysis, is Large Eddy
Simulations (LES) in which only the eddies of smaller scales are modelled, whereas the larger eddies
scales are numerically resolved using statistical techniques based on the Kolmogorov energy cascade
analogy [14]. High fidelity simulations are usually associated with Actuator Line Models (ALM)
considered to be more accurate in the calculation of body forces at the rotor [11].
Another important factor influencing turbine wakes is the atmospheric characteristics at the site and
their change with time. The most commonly used wake models do not take atmospheric stability into
consideration. However, some commercial wake models currently used in the offshore wind market
such as FUGA [14] and FarmFlow [15], do use some simplified methods to estimate the effects of
stability on the wake development. Deliverable 1.4 [16] of the CL-Windcon [17] project shows a
comparison between different engineering wake models against a field test of a utility-scale wind turbine
in different atmospheric conditions.
The lack of atmospheric stability corrections in common wake models used for long-term energy
yield calculations can easily be explained: taking the average of a sufficiently long time of turbine
operation, the non-neutral atmospheric conditions average out to give neutral atmospheric conditions,
at least for sites where very stable or very unstable conditions are not predominant. What is more, the
estimation of the occurrences of different stability characteristics at the site can be a difficult task, since
requiring the measurement of specific wind characteristics.

1.1 Objectives and outline

The main objective of this paper is to propose a new approach for the wake modelling of predominantly
stable or unstable atmospheric conditions, detailed in Section 3. The proposed model is a modification
of the Ainslie model [4], taking the effects of atmospheric stability into account. This model does not
introduce any additional experimental parameter hence not requiring additional calibration against
experimental data, and only requires an estimate of the Monin-Obukhov length (or MOL) [14]. A
validation of this model against a set of experimental and wind farm production data (detailed in
Section 2) is also presented, and an overview of the methodology for atmospheric stability classification
is provided. In addition to the above, another objective of this study is to increase confidence in the use
of existing wake deflection models for the purpose of wind farm control and optimisation.
It is also highlighted that this paper will only focus on simplified engineering models and steady-
state turbine wakes, therefore not considering dynamic affects such as wake meandering. All the tests
presented in this paper have been performed using LongSim, a software created and used internally in
DNV GL for advanced studies on wind farm steady-state and dynamic simulation, set-point optimisation
and wake steering optimisation.

2. Experimental and numerical data used for the calibration of models

A set of different datasets have been used to perform wake model calibrations and validations. These
are detailed in the following sections.

2.1 Wind tunnel measurements and setup

Wind tunnel experiments were carried out at the Politecnico di Milano between 2017 and 2018, as part
of the CL-Windcon project [17]. The wind tunnel is a closed loop, boundary-layer type and different
turbulence intensities and wind shear profiles can be reproduced. A turbine model with a rotor diameter
of 1.1 m, named G1, was designed and built (a detailed description of the turbine model can be found
in the public Deliverable no. 3.1 [18]). Experiments were carried out for an isolated turbine and for
groups of turbines, but only the single-turbine results have been used in this paper. Free flow wind speed

2
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

at the turbine’s hub height (0.825 m or 0.75 rotor diameters) is measured using a Pitot tube located 3
rotor diameters upstream of the turbine, and a full inflow-plane investigation was previously carried out
to characterise the free flow (available in public Deliverable 3.4 [19]). Hot-wire probes were used to
record flow measurements (with a sampling frequency of 2 kHz) which were averaged over a minute to
obtain steady-state effects at 5, 7.5 and 10 rotor diameters behind the turbine location. All the tests were
carried out with a mean wind speed of 5.5 m/s.
Two different flow conditions were reproduced in the wind tunnel experiments: the offshore
condition is characterised by an averaged wind shear profile corresponding to a power law exponent of
0.08 and lower flow-wise turbulence, whereas the onshore wind shear condition corresponds to a power
law exponent of 0.20 and higher flow-wise turbulence (a full description and characterisation of the
wind tunnel experiments is presented in the public Deliverables [18, 19]). The standard deviation of the
flow-wise free-stream wind speed component 𝜎𝑈 is shown in Figure 1, measured at the wind tunnel test
section with no turbine models installed and measured at the location of the turbine rotor plane. The
measured 𝜎𝑈 remains almost constant with height for the offshore case whereas it shows a noticeable
decrease (for increasing height) for the onshore case across the rotor area. Turbulence stochastic models
[20] generally assume that 𝜎𝑈 is constant with height. Although it is suggested to use the value of
turbulence intensity averaged across the rotor, it is anticipated that this might cause some discrepancies
between the wake models and the measured wake profiles.
Flow measurements were carried out both for a zero-yaw configuration and for different yaw angles,
spanning from -40° to 40° with a 10° step. Detailed investigations have also been carried out on the
measured time-series at different locations relative to the wind turbine, in particular to analyse the energy
spectra, the estimated integral length scale and the measurement scattering. The spectral analysis
(performed using a Welch approach with the help of a Hanning filter and a de-trending filter) has not
shown unexpected features and the turbulent dissipation rate of the inertial subrange seems to agree well
with the Kolmogorov law. These results are not shown here, but indicate that the turbulent flow
development within the wake behaves as expected and shows no unexpected artefacts in the
measurements provided, which are deemed of good quality.

Figure 1. Variation with height of the standard deviation of the horizontal wind speed component for
both offshore and onshore case, measured for the characterisation of the free-stream conditions in the
wind tunnel at Politecnico di Milano.

2.2 CFD simulations

A set of time-averaged results from the open-source SOWFA computational solver [12] was provided
by the National Renewable Energy Laboratory (NREL). SOWFA uses a detailed Actuator Line Model
(ALM) to produce high-fidelity unsteady simulations of wakes and turbine loads. The simulation
entailed a single prototype 5MW turbine with a rotor diameter of 126 m and a hub height of 90 m, with
averaged free-stream wind shear profile with power law exponent of approximately 0.18, averaged inlet

3
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

wind speed of 8 m/s and turbulence intensity estimated to be approximately 10 % (therefore

characterised by overall neutral flow conditions). The inflow conditions are generated in a precursor
simulation run using a one-equation eddy-viscosity model for the definition of the subgrid scale,
therefore the averaged wind characteristics can show variations in the free-flow.

2.3 Wind farms production data

For the purpose of comparing the proposed wake model to other wake models and test its applicability
to a real case scenario, production data from one small onshore wind farm, Wieringermeer, and two
large offshore wind farms, Horns Rev I and Nysted, were utilised.
The Wieringermeer Wind Farm site consists of 5 wind turbines installed in a test site in Netherlands.
These are variable speed and pitch turbines of 80 m diameter, 2.45 MW rated power and hub height of
80 m. TNO provided the data as part of the UPWIND project and gave permission to DNV GL to use
mast measurements and power turbine production data (for a total of approximately 3.2 years of
available data) at a single turbine, namely T6: this is the second turbine from the west, being waked by
turbine T5 for westerly winds. The atmospheric conditions at the site were measured at a mast located
in the vicinity of the turbines and were made available for wind directions ranging approximately
between 230° and 320°. The instruments installed on the mast include a thermocouple measuring the
temperature difference between 37 m and 10 m heights, 3 sonic anemometers and 3 cup anemometers
mounted at heights of 108 m, 80 m and 52 m, and two wind vanes at 79.2 m and 51.2 m. Further details
about the on-site measurements are provided in [21]. As the individual pressure and temperature signals
were not provided, these were retrieved from the ERA5 [25] reanalysis dataset at the node centred
approximately 4 km to the south of the mast. Although the uncertainty given by such dataset are
considered larger than the ones associated to direct site measurements, these are considered acceptable
for the purpose of the estimation of the atmospheric stability at the site. The results presented in this
paper show binned mast and turbine data using 2° wind direction bins, as well as filtering out wind speed
outside the range 6-8 m/s. Also, as the purpose of this analysis is to compare the model to measured data
where highly stable and unstable atmospheric conditions are present, only the summer months (June to
August) were considered, and the dataset has been subdivided into daytime and night-time periods,
respectively ranging between 7-18 hours and 20-5 hours.
Horns Rev I and Nysted are large offshore wind farms situated respectively 20 km off the west coast
of Denmark and 15 km off the south coast of Lolland Island, Denmark. Details on the turbine model,
hub height and inter-turbine spacing are detailed in Table 1. Both wind farms have an oblique rectangle
layout as shown in Figure 2. The atmospheric stability conditions at the sites have been classified based
on 12 months of data recorded at the Horns Rev I site masts and based on 3 years of data recorded at the
Nysted site mast: these were for the UPWIND project [22] from the calculation of the Richardson
number based on the temperature difference and wind speed difference at two heights. The stability
classification for Horns Rev I is subdivided into seven categories (very unstable, unstable,
neutral/unstable, neutral, neutral/stable, stable, very stable), whereas in Nysted six categories were used
(very unstable, unstable, neutral/unstable, neutral/stable, stable, very stable): these are shown in
Figure 2. The production data and the averaged wind conditions (mean wind speed, wind direction and

Horns Rev I Nysted

Turbine model Vestas V80 2.0 MW Bonus 2.3 MW
Diameter [m] 80 82.4
Hub height [m] 70 68.8
10.3 east-west
Inter-turbine spacing [D] 7
5.8 north-south
2 hours of data for each wind
Available SCADA data 18 hours of data
direction and wind speed bin
Directional bins 255° - 285°, 5° step 263° - 293°, 5° step
Reference turbine T07 A5
Table 1. Summary of characteristics of Horns Rev I and Nysted wind farms.

4
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

Figure 2. Layout of the Horns Rev I (a) and Nysted (b) offshore wind farms, and respective histograms
of the occurrence of different stability classes (c.d) for the 10 m/s wind speed bin. Layout of the
Wieringermeer wind farm (e). Reference turbines in red.

mean turbulence intensity) recorded at the sites have been provided for three wind speed bins (6±1 m/s,
8±1 m/s and 10±1 m/s) and 7 direction bins as detailed in Table 1. The power production data for each
turbine was provided as normalised to the reference upwind turbine for each respective site. Details on
the data used for the modelling of the wind farms’ production are given in Section 4. Production and
wind condition data for both Horns Rev I and Nysted wind farms have been provided as part of the
UPWIND project [22, 23, 24].

5
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

3. Wake models
3.1 Wake deficit model
The Ainslie model is part of that class of turbine wake models called Gaussian models and it is
sometimes referred to as eddy viscosity model (or EVM). It is obtained solving the Reynolds-Averaged-
Navier-Stokes (RANS) equations in two dimensions and formulated in cylindrical coordinates, using
the assumptions for incompressible, stationary and axisymmetric flow and neglecting gravity forces and
pressure gradients. By using the axisymmetric flow and the Bousinnesq assumptions, the viscous forces
can be simplified to
𝜇 𝜕𝑢𝑥 𝜕𝑢𝑥
−𝑢𝑥 𝑢𝑟 = =𝜈 (1)
𝜌 𝜕𝑟 𝜕𝑟

In equation (1), 𝜌 is the air density, 𝜇 is the dynamic viscosity of the air, x and r are respectively the
axial and radial coordinates, and 𝑢𝑥 and 𝑢𝑟 are respectively the axial and radial velocity components.
Also, the kinematic viscosity (or momentum diffusivity) indicated as 𝜈 in the equation above can be
expressed as 𝜈 = 𝜈amb + 𝜈𝑡 , where the first is the ambient kinematic viscosity (small for air in standard
conditions, in the order of 1.5 · 10-5 m2/s and usually neglected), and where 𝜈𝑡 is the eddy viscosity term.
Based on dimensional analysis, this term can be rewritten as: 𝜈𝑡 = 𝑘1 𝑙𝑤 𝑈𝑤 + 𝐾𝑚 , where 𝑙𝑤 is assumed
to be of the order of the wake width (as defined by Ainslie, roughly two times the full-width-at-half-
height, or FWHH, of the Gaussian self-similar wake shape), 𝑈𝑤 is a velocity scale, 𝐾𝑚 is the term for
the ambient momentum diffusivity and 𝑘1 is a constant which value, used throughout this study, is based
on extensive internal validation from DNV GL. Using the assumption of isotropic turbulence, the
𝜅𝑢∗ 𝑧
ambient momentum diffusivity can be written 𝐾𝑚 = 𝑧 where 𝜅 = 0.41 (von Karman constant), Φ𝑚
Φ𝑚 ( )
𝐿
is the stability correction term (as a function of height, z, and Obukhov length, L) and 𝑢∗ = √𝜏𝑖𝑗 /𝜌 is
the friction velocity, where 𝜏𝑖𝑗 indicates the shear stress. For a non-neutral flow, the friction velocity
can obtained from the following equation of the non-neutral wind profile:

𝑢∗ 𝑧 𝑧
𝑈= ln ( + 𝛹𝑚 ( )) (2)
𝜅 𝑧0 𝐿
with z0 being the roughness length. The function Ψ𝑚 is zero for neutral conditions and it can be defined
as follows, according to Businger-Dyer [26] and Högström [27]:
𝑧 5𝑧
• Stable conditions: Ψ𝑚 ( ) = − (3)
𝐿 𝐿
𝑧 1 + 𝑥2 1 + 𝑥 2 𝜋
• Unstable conditions: Ψ𝑚 ( ) = ln [( )( ) ] − 2 𝑎𝑟𝑐𝑡𝑎𝑛 𝑥 + (4)
𝐿 2 2 2
where
1
19.3𝑧 4 (5)
𝑥 = (1 − )
𝐿
We can rewrite the whole eddy viscosity term as follows, after normalising by 𝑈HH 𝐷 (where D is the
rotor diameter):

𝐻𝐻
𝜈𝑡 𝑏𝑤 𝑈𝐶 𝜅2 ( )
= 𝑘1 ( ) (1 − )+ 𝐷
𝑈𝐻𝐻 𝐷 𝐷 𝑈𝐻𝐻 𝐻𝐻 𝑧 𝑧 (6)
ln ( 𝑧 + Ψ𝑚 (𝐿)) Φ𝑚 (𝐿)
0

6
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

In the formula above, 𝑏𝑤 is the wake width and 𝑈𝑐 is the wake centreline velocity. Also, note that no
assumption that 𝐻𝐻 ≈ 𝐷 is used and the stability correction term Φ𝑚 is defined in the next section.
Also, the filter function used by Ainslie for near-wake regions is set to F=1, hence this formula is valid
𝑢
only in the far-wake. Furthermore, using the approximation 𝜅∗ ≈ 𝜎𝑢 [23, 28], the logarithmic function
above can be approximated to 1 / TI, where TI indicated the turbulence intensity. The proposed model
will be referred in this paper as the Stratified-EVM wake model.

3.2 Experimental laws for atmospheric stability

Different experiments were carried out to describe the wind and temperature profiles in the atmospheric
boundary layer (ABL), most notably the Kansas study [20], from which the Businger-Dyer experimental
model was obtained [26], and some tall mast experiments [29] from which the Högström [27] and the
Gryning [30] models originate. The Businger-Dyer relationship (3,4) is used in this study with the
revised parameters by Högström. This defines the correction term Φ𝑚 as follows:

5𝑧
• Stable conditions: Φ𝑚 = 1 + 𝐿
(7)

19.3𝑧 −0.25
• Unstable conditions: Φ𝑚 = (1 − ) (8)
𝐿
The set of equations above depend on the Monin-Obukhov length, L (or the Monin-Obukhov stability
parameter 𝑧/𝐿). This parameter can be difficult to estimate, as it changes throughout the day (and
therefore should be seen rather as a distribution or a time-series) especially for sites with extreme
weather conditions. Another common parameter used for the classification of stability in the atmosphere
is the bulk Richardson number, which can be defined as:
𝑔(𝜃̅ − 𝑇𝑠 )Δ𝑧
𝑅𝑖𝐵 = (9)
𝑇𝑠 Δ𝑈 2
where g is the gravitational acceleration constant, 𝜃 is the potential temperature, which can be defined
𝑅𝑑
−( )
𝑃 𝐶𝑝
as 𝜃 = 𝑇 (𝑃 ) , where 𝑅𝑑 is the gas constant for dry air and 𝐶𝑝 is the specific heat constant for
𝑟𝑒𝑓
air, P is the atmospheric pressure, 𝑃𝑟𝑒𝑓 the pressure reference and 𝑇𝑠 the temperature near the ground
expressed in Kelvin.
The ranges of MOL, Obukhov stability parameter and bulk Richardson number typically used in the
industry are summarised in the following table.
Stability class Pasquille [31] L [m] [14] z / L [-] [32] RiB [33]
Very unstable A -100 < L < -50 RiB ≤ -0.023
Unstable B -200 < L < -100 -1.4 < z/L ≤ -0.35 -0.023 ≤ RiB < -0.011
Neutral/Unstable C -500 < L < -200 -0.011 ≤ RiB < -0.0036
Neutral D |L| > 500 |z/L| < 0.35 -0.0036 ≤ RiB < 0.0072
Neutral/Stable E 200 < L < 500 0.0072 ≤ RiB < 0.042
Stable F 50 < L < 200 0.35 ≤ z/L < 7.0 0.042 ≤ RiB < 0.084
Very stable G 10 < L < 50 RiB ≥ 0.84
Table 2. Summary of atmospheric stability classes used in literature, expressed using different metrics.

3.3 Turbine interaction

Every turbine affected by the wakes of a turbine upstream will experience inlet flow characteristics
which cannot be approximated to the ones experienced in the free, undisturbed flow. Two main

7
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

characteristics are taken into consideration when modelling the wake at a turbine downstream: the added
turbulence model and the superposition model.
The added turbulence model allows to estimate the turbulence intensity seen by the affected turbine.
Two approaches are typically used: the Quarton-Ainslie model [34] and the Crespo-Hernandez model
[35]. They model the turbulence intensity at the affected turbine as the combination of the free-stream
turbulence intensity and an added turbulence, which is modelled as a function of turbulence intensity
and thrust coefficient (in the case of the first model) or as a function of turbulence intensity and axial
induced velocity (in the case of the second model). Moreover, it is noted that the Crespo-Hernandez
model is considered valid only in the far-wake, assumed to start at 5 rotors diameters downstream of the
turbine, whereas this limitation is not considered in the Quarton-Ainslie model.
Different superposition models are commonly used in order to obtain combined effects from
superposition of turbine wakes, hence predicting combined wake velocity deficits and wake added
turbulence. Examples of these models are dominant wake and sum-of-deficits models, however none of
it has been shown to have a clear advantage in the literature (see [36, 37] for more details), and certainly
none of these are based on proper physical laws, but rather mathematical or pragmatic laws.
The models for added turbulence intensity and turbine wake superposition are not discussed and
validated in depth in this publication. However, it is worthwhile to stress how these models are
fundamental for the overall correct modelling of wind-farm-wide wake effects. For instance, when
additional deep-array effects are introduced, such as the large wind farm corrections used in Windfarmer
[38], materially different predictions can be obtained when different superposition models are used. It
is also underlined that different assumptions on added turbulence affect the wake dissipation of
downstream turbines, which is important to keep in mind when comparing different models: for instance,
the increased wake dissipation downstream due to the added turbulence is not considered in the model
from Bastankhah & Porté-Agel [5], contrarily to the wake model used in Windfarmer.

3.4 Models for yawed rotors

The theory behind the flow characteristics for a yawed rotor has roots in the helicopter industry since
used to calculate the rotor loads during forward flight. Glauert’s correction for yawed rotors to the
Rankine-Freude actuator disk model [39] and Coleman’s study on skew deflection angle in the rotor
near-wake (based on a vortex line approach) [40] are the main examples. More recently, two theories
have been extensively used in literature to model the wake past yawed rotors: the Jimenez model [41]
and the model proposed by Bastankhah & Porté-Agel (commonly referred to as the EPFL model, also
in this paper) [5,6]. It is worth mentioning that other, more recent studies have appeared in literature,
trying to study the wake deflection effect from a vorticity point of view [42, 43], but these lie outside
the scope of this paper. The Jimenez model is based on a momentum balance approach and does not
take into account the tangential losses caused by the three-dimensional characteristics of the wake flow,
as well as any effect from atmospheric stability; it is therefore considered to overestimate the wake
deflection. Moreover, it uses one empirical parameter which needs to be calibrated against
measurements, even if a link between this parameter, 𝐾𝑑 , and the turbulence intensity has been suggested
in [28].
The EPFL model uses two Gaussian distributions aligned to the vertical and lateral axes (with respect
to the plane normal to the flow direction) to allow for different velocity deficit distributions in the two
axes. After approximating the wake growth rate within the far-wake region as linear and estimating the
length of the potential core, used to define the inlet boundary conditions for the far-wake region (based
on Coleman’s approximations for the wake skew angle), it is possible to calculate the lateral wake
deflection due to a given yaw angle, depending on both the incoming turbulence intensity and the thrust
coefficient. The EPFL model needs two empirical parameters for the wake velocity deficit calculations
and two more for the lateral deflection calculations: these were estimated from wind tunnel experiments
and LES simulations [5]. However, more validation on real-scale turbines would be beneficial to obtain
reliable sets of empirical parameters to be used for different classes of turbines. Furthermore, it is worth
commenting on the importance of the definition of the velocity and thrust vectors in the different models,

8
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

which might differ from one another. For example, in the Ainslie model, the wind speed is defined as
perpendicular to the rotor plane, and this will need to be taken into account when calculating the thrust
coefficient for a yawed rotor, as the final formula will depend on which direction the velocity and the
thrust force vectors are projected onto.
Additionally, the effects of wake centreline displacement due to the combined effect of wake rotation
and non-uniform inflow shear have been described in [44] and these are discussed in the test cases used
for this study.

4. Results and discussion

4.1 Wind tunnel tests
4.1.1 Wake velocity deficit comparisons (zero yaw) for an isolated turbine.
Comparison between the measured, steady-state wake velocity deficit plots and the wake model
predictions are performed at a horizontal plane at hub height (corresponding to 0.75 rotor diameters).
The standard Ainslie model, the EPFL model and the Stratified-EVM model are compared to the
measurements at the three downstream planes, as shown in Figures 3 to 5.
In Figure 3, the standard Ainslie model is shown not to fit the experimental data. The EPFL model
is compared to the wind tunnel measurements in Figure 4. A considerable difference between the plots
labelled (a,b) and the ones labelled (c,d) is visible. By using the EPFL parameters originally obtained
by Bastankhah & Porté-Agel [5], the wake model manages to fit reasonably well the measurements with
low turbulence (a), whereas it performs poorly in the other test case with higher turbulence intensity (b).
A new set of parameters was optimised from the measurements using a cost-minimising function
approach during the CL-Windcon project [16]. As it can be seen in plots (c, d) this new set of parameters
improves the model fit noticeably. Although the optimisation of the set of parameters could constitute a
good mathematical approach to calibrate the wake model for any specific case, it could be unfeasible
when used for wake assessments where wake measurements are not available. The two sets of
parameters are shown in Table 3.
𝛼 𝛽 𝑘𝑎 𝑘𝑏
Proposed in [5] 2.32 0.154 0.3837 0.003678
Optimised in [16] 1.2973 0.2375 0.1177 0.0194
Table 3. Sets of parameters used in the EPFL model, used for the wind tunnel comparisons

Figure 3. Standard Ainslie model applied to the wind tunnel tests, both for the offshore (a) and onshore
(b) flow case.

The Stratified-EVM wake model is compared to the same test case, as shown in Figure 5. Based on the
broad analogies between the inflow conditions and the different stability classes in the literature (see

9
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

Table 2), different Obukhov lengths were used to fit the measurements. For the offshore case,
characterised by a low turbulence intensity and predominantly stable conditions, a MOL of 200 m has
been found to give the best fit with the measurements at the most upstream plane. An Obukhov length
characteristic of stable conditions, however, results in the overprediction of the wind speed deficits at
the two most downstream measurement planes, for which neutral characteristics were imposed to obtain
a reasonable fit with the data, using a MOL of 2000 m. For the onshore case, instead, using a MOL of
450 m representative of slightly stable conditions allows to obtain a reasonable fit to the experimental
data at the three measurement planes, with a small wind speed deficit underprediction for the most
upstream measurement plane and a larger wind speed deficit overprediction for the most downstream
measurement plane. As it can be observed comparing the onshore and the offshore test cases, the wake
centreline wind speeds at the two more downstream planes are similar. This might suggests that the
wake in the offshore case tends to recover quickly to the same wind speeds observed in the onshore
case, which might explain the need to use different Obukhov lengths for different measurement planes
in order to obtain a good fit for this test case. This might also explain the not satisfactory fit found using
the standard Ainslie model for both the offshore and the onshore cases.
It is noted that the MOL values mentioned above are kept as an approximated round number, and a
best fit value is not calculated using optimisation techniques in order to avoid overfitting. Due to the
nature of the wind tunnel tests, and hence the lack of proper atmospheric stability classification, it is not
possible to obtain a precise value for the MOL. Moreover, as the atmospheric boundary layer and the
turbine model are scaled down in this experimental setup, the MOL has been also scaled down by a
factor of 100, under the assumption that the turbine model in question is the (approximately) 1:100 scale
reproduction of a utility-scale turbine.
In conclusion, the uncertainties associated with the behaviour of the flow and the atmospheric
stratification characteristics recreated in the wind tunnel do not allow to use this data to fully validate
the model. Nevertheless these tests show that when an Obukhov length in line with the indicative
atmospheric characteristics recreated in the wind tunnel is chosen to modify the eddy viscosity term in
the Ainslie model, this can be used to obtain a satisfactory prediction of the turbine wake. In order to
gain more confidence in the proposed model, this will be compared against additional test cases in this
paper, including utility-scale operational wind turbines.

4.1.2 Wake centreline deflection due to rotor yaw, for an isolated turbine
The measurements for the non-zero yaw cases were analysed and compared to the Jimenez and EPFL
models, as detailed in Section 2. As described above, it is known in the literature that the combination
of a rotating turbine wake and a non-uniform wind shear profile deflect the wake centreline even when
the turbine’s yaw angle is zero. Analysing the wind tunnel measurements, no clear indication of such
deflection was found and therefore it is not considered in the results presented in this section.
Different 𝑘𝑑 parameters were tested in order to obtain a closer fit from the Jimenez model to the
wind tunnel measurements. A value of 0.1 was deemed to be the optimal value. This value is lower than
the other ones found in literature for utility-scale wind turbines [8, 41], which has the effect of increasing
the centreline deflection. When using the EPFL model, the four optimised parameters for this wind
turbine model test will be used for these comparisons. The measurements are compared to the Jimenez
and the EPFL models in Figure 6 for negative yaw angles. The Jimenez model better fits the
measurements for the onshore test case, whereas the largest discrepancy is obtained for the largest
absolute yaw angles in the offshore test case. A possible explanation to this could be the lack of any
relationship to the variation of atmospheric stability within this model: as the flow becomes more stable,
the overall wake deflection will be larger due to the reduced effect of flow mixing and entrainment along
the wake. This indicates that in order to obtain a better fit for specific atmospheric conditions using the
Jimenez model, specific 𝑘𝑑 parameter should be chosen for each case, which might be unpractical. The
EPFL model shows smaller deflection discrepancies compared to the Jimenez model: this is particularly
true for the offshore test case. Although a large discrepancy is noticeable for the onshore test case at a

10
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

Figure 4. EPFL model applied to the wind tunnel tests, both for the offshore (left) and onshore (right)
flow case. The set of parameters proposed in [5] have been used in (a-b), whereas the set of parameters
proposed in [16] are used in (c-d).

Figure 5. The Stratified-EVM wake model applied to the wind tunnel tests, both for the offshore (a)
and onshore (b) flow case.

11
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

yaw angle of -40°, this is considered to be of lesser importance at least as far as wake steering
applications are concerned, as smaller angles are usually used due to the risk of large increases in loads.

Figure 6. Wake lateral deflection due to yawed rotor for the wind tunnel test case, comparing Jimenez
and EPFL models. (a,b,c,d) are for the offshore case, respectively at -10°, -20°, -30°, -40°; (e,f,g,h) are
for the onshore case, respectively at -10°, -20°, -30°, -40°.

Although both models compare reasonably well to the measurements, it is highlighted that additional
validations are needed to increase confidence on the parameters to be used for real-world applications,
in both models.

12
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

4.2 CFD simulations

In this section, data from CFD simulations at different yaw angles are compared against the two Jimenez
and EPFL wake deflection models. It is worth underlying these CFD results are not representative of
any field test and are not validated against actual field measurements. Therefore, the CFD data analysed
here are used assuming they realistically represent the wake development behind the turbine.

4.2.1 Wake velocity deficit comparisons

The simulated wake deficit profiles at four different downstream planes are initially compared to the
proposed wake model (Figure 7a) and the standard Ainslie model (Figure 7b), for the turbine simulation
with no rotor yaw. Based on the comparisons at four downstream planes (4, 6, 8 and 10 rotor diameters),
using a MOL of -1000 m (characteristic of neutral conditions, see Table 2) allowed to obtain a good fit
between the proposed model and the simulated wake deficits. The centreline wake deficits from the
simulation and the two wake models are shown in Figure 7b: both models show to predict the wake
deficits reasonably well.

Figure 7. Wake velocity deficit comparisons for the CFD test case, shown at 4, 6, 8 and 10 rotor
diameters downstream: (a) Stratified-EVM wake model with MOL=-1000 m, (b) Standard Ainslie
model and Stratified-EVM wake model.

It is visible from Figure 7 how the centreline of the velocity deficits tends to shift towards the left for
increasing distance downstream: this is caused by the combined effect of the rotating turbine wake and
the non-uniform inflow wind shear profile. In the next section, it is shown how the knowledge of this
component of wake displacement allows to obtain better wake deflection predictions.

4.2.2 Wake centreline deflection due to rotor yaw, for an isolated turbine
Both the Jimenez and the EPFL models were used to predict lateral wake displacement due to yawed
rotor for this CFD test case in neutral flow. Simulations of the same turbine analysed above have been
carried out at the following yaw angles: -25°, -20°, -15°, 10°, 15°, 20°, 25°.
The parameters originally proposed by Bastankhah & Porté-Agel [5] have been used in these tests
for the EPFL model, whereas different values of kd have been used in the comparisons in Figure 8 and
Figure 9. In order to take into account the wake centreline deflection for non-yawed rotor visible in
Figure 7a, the centreline deflections from the LES data were linearly summed to the deflection
component obtained for the no-yaw case, an approach which is also used in the software FLORIS [9,
10]. The plots in Figure 8 and 9 show both the deflections obtained from the LES data (indicated simply
as ‘LES’ in the plots) and the summed deflection components (indicated with ‘LES + Linear deflection’).

13
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

Figure 8. Wake lateral deflection for the CFD test case, shown for yaw angles -25°(a), -20° (b), -15°
(c) and including the EPFL model and the Jimenez model (with different kd parameters).

The plots clearly show that if the no-yaw component of centreline deflection is not taken into account,
the EPFL model largely overpredicts (for negative yaw angles) and underpredicts (for positive yaw
angles) the wake deflection. The Jimenez model is shown to be able to better approximate the wake
deflection downstream of the turbine, however very different values would need to be used for the k d
parameter for positive and negative angles.
When the effect of the non-yaw deflection component is removed via linear sum, it can be seen how
both the EPFL model and the Jimenez model better approximate the data from the simulations.
Particularly for positive yaw angles, both the EPFL model and the Jimenez model (with a kd parameter
of 0.15) predict well the wake deflection downstream of the turbine for all the angles used in this
investigations. When the yaw angles are negative, an acceptable fit of the two models with the data is
visible only for the -25° and -20° yaw angle cases and up to approximately 5 rotor diameters downstream
of the turbine.
These comparisons show how both the models used for this test case can predict reasonably well the
wake deflection in neutral conditions, although the centreline deflection for no-yaw needs to be
estimated to give reliable results. To the best knowledge of the authors, no low-fidelity or medium-
fidelity models are available in the literature to help predict this deflections effect.

4.3 Wind farms production data

In order to further increase confidence in the described methodology, wind farm production data is used
in this section to perform comparisons with the proposed model.

14
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

Figure 9. Wake lateral deflection for the CFD test case, shown for yaw angles 25°(a), 20° (b) 15° (c)
10° (d), including the EPFL model and the Jimenez model (with different kd parameters).

4.3.1 Wieringermeer onshore wind farm

This dataset was used to test the Stratified-EVM wake model against on-site meteorological mast data
and SCADA data at a site where the effects of atmospheric stability and instability are not negligible.
The power production data for Turbines T06 are normalised by the production of the leading turbine for
westerly winds, T05. The data is binned for the specific wind speed range 6-8 m/s and was also binned
for wind directions every 2°, between 230° and 320°. As a first test, the whole production dataset was
compared to the proposed wake model using a range of MOL values characteristic of neutral
atmospheric conditions: as it can be seen in Figure 11(a), a good fit is obtained using an MOL ranging
from 750 m to 1000 m, and the results from the Windfarmer model are also superimposed for reference.
In order to compare the model to measured data where highly stable and unstable atmospheric conditions
are present, only the summer months (June to August) were considered, and the dataset has been
subdivided into daytime and night-time periods (respectively ranging between 7-18 hours and 20-5
hours). It is expected that the summer-daytime atmospheric characteristics exhibit signs of unstable
conditions, whereas the summer-night-time atmospheric characteristics exhibit signs of stable
conditions. In order to estimate the on-site atmospheric stability, the bulk Richardson number was
calculated for each 10-minute averaged record from the mast signals and its distribution is shown for
both daytime and nigh-time in Figure 10. Binned wind speeds, binned wind directions and binned
turbulence intensities were used as model inputs for each directional bin and were compared to the
measured, filtered data. The comparison shows how the daytime and night-time filtered production data
at Turbine T06 exhibit different wake effects due to the predominance of unstable and stable

15
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

atmospheric effects, respectively. The median bulk Richardson number for each distribution has been
calculated and empirical formulas have been used to relate them to the correspondent Obukhov lengths,
as described in Holtslag [29]. The results obtained from the Stratified-EVM wake model are shown in
Figure 11(b): the model appears to be able to correctly predict the power deficit due to wake of the
preceding turbine for both predominantly stable and unstable conditions.

Figure 10. Distribution of bulk Richardson number for daytime(a) and night-time (b), both for the
summer period, at the Wieringermeer site.

4.3.2 Horns Rev I and Nysted offshore wind farms

Turbine production data at the Horns Rev I offshore wind farm, provided as part of the UPWIND project
[22], have been used to further validate the proposed Stratified-EVM wake model against a real-world
scenario. In order to model the production for all the wind turbines of the wind farm, the choice of the
added turbulence model and superposition model required particular attention. The Crespo-Hernandez
model [35] was shown to perform well on previous wake model validations based on a large offshore
wind farm, like in [6]. Since preliminary tests indicated this added turbulence model performed better
than the Quarton-Ainslie model for the specific test setup, and since the relatively large inter-turbine
spacing at the Horns Rev I wind farm falls within the indicated range [35], this model was selected for
this test case. Both the dominant wake and the sum-of-deficits superposition models have been tested
on this wind farm validation exercise and the latter has shown to have a noticeable improvement in the
turbine row power predictions (this is also in accordance to what is shown by Niayifar & Porté-Agel
[6]). Based on the available information on the stability classification at the site and the occurrence of
each stability class, the Stratified-EVM wake model was run using three different MOL setting and the
results were weighter-averaged according to the neutral, stable and unstable occurrences at the site, as
shown in Figure 2. The measured power production for each turbine row is compared to the modelled
power production in Figure 12(a). The comparison is shown for the specific 10±1 m/s wind speed bin.
The same added turbulence model and wake superposition model used for the Horns Rev I wind farm
above, were applied to the Nysted offshore wind farm too. Measured power production for each turbine
row and representative of the 10±1 m/s wind speed bin is compared to the modelled power production
in Figure 12(b). For this specific test case, the results are the weight-averaged of two runs of the
proposed model, to better follow the provided stability classification at the site (i.e. one run using an
Obukhov length representative of all the stable and neutral/stable subclasses, and another using an
Obukhov length representative of all the unstable and neutral/unstable subclasses).
The power production predicted at the two sites using the DNV GL’s Windfarmer wake model is
also shown for comparison in Figure 12, which fits the production data better than the proposed model.

16
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

Figure 11. Wieringermeer: power production at Turbine T06, normalised by power at turbine T05,
shown for filtered and binned data. (a) neutral case, (b) stable and unstable cases.

Nevertheless, the proposed model is deemed to perform reasonably well especially for the first turbine
rows. The largest discrepancies between the production data and the proposed model are found in the
last turbine rows in both sites. It is noted that the turbine production data and the atmospheric
classification are provided as averaged values, therefore it was not possible to make a time-series-based
comparison by filtering and isolating stable and unstable occurrences to be compared against the model
(as it was done for the Wieringermeer test case). As the greatest discrepancies are visible in the last wind
farms’ rows, it is suspected the choice of a specific wake superposition model can heavily influence the
results, and it is therefore considered by the authors as an area where further investigations are needed.

5. Conclusions
A wake model based on a formulation of the Ainslie model for stratified flow has been proposed in this
work: the eddy viscosity parameter characterising the Ainslie model was modified to take into account
the effects of atmospheric stability. This model has been initially calibrated against wind tunnel tests
carried out for two different inflow conditions (characteristic of neutral onshore and stable offshore

17
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

scenarios), and against the results from an LES simulation of a utility-scale wind turbine in neutral
atmospheric conditions. The proposed model showed to be able to predict reasonably well the wake
deficits for these test cases, once an appropriate Obukhov lengths characteristic of specific atmospheric
stability conditions is selected. Due to the nature of the wind tunnel tests and the LES simulations, it
was not possible to obtain an estimate of the Obukhov length from the data. Nevertheless, it is shown
that when a value of Obukhov length is chosen in line with the indicative atmospheric characteristics of
each test case, hence modifying accordingly the eddy viscosity term in the Ainslie model, this can be
used to obtain a satisfactory prediction of the turbine wake.

Figure 12.s Measured and modelled turbine power averaged for each turbine row, normalised by
reference power. (a) Horns Rev I wind farm, (b) Nysted wind farm.

Production data from three wind farms have also been used to increase confidence in the validity of
this model. The Wieringermeer wind farm was used to further validate the model for neutral and non-
neutral atmospheric conditions. The proposed model was calibrated against the whole production dataset
characterised by neutral atmospheric conditions, and a MOL ranging between 750 m and 1000 m
allowed to predict well the production at the test turbine. Furthermore, the measured production data
and the meteorological data were further filtered to isolate daytime and night-time data during the
summer months. The bulk Richardson number was estimated using the filtered data and a correspondent
Obukhov length was obtained for both time periods. The use of a characteristic Obukhov length on
meteorological measurements for both time periods allowed to obtain a good agreement between
modelled and measured power production.
In order to test the proposed model against large wind farms, production data from two large offshore
wind farms (Horns Rev I and Nysted) were used. The proposed model was used alongside the Crespo-
Hernandez model for added turbulence and the sum-of-deficits wake superposition model. The stability
classification at these sites was available as probability of occurrence for each considered wind speed
and direction bin, therefore a weighted average approach was used to combine simulations
representative of neutral, stable and unstable conditions, using representative Obukhov lengths. The
proposed model is deemed to fit reasonably well to production data especially for the initial turbine
rows, although the wake model used in DNV GL’s Windfarmer is shown to better fit the production
data. As the greatest discrepancies are visible in the last wind farms’ rows, it is suspected the choice of
a specific wake superposition model can heavily influence the results, and it is therefore considered by
the authors as an area where further investigations are needed.
Additionally, the EPFL and the Jimenez models for wake deflection have been compared using both
the wind tunnel and the high-fidelity CFD simulation test cases. For the wind tunnel experiments, which

18
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

were carried out for two different flow conditions, the EPFL model was found to better predict the wake
deflections compared to the Jimenez model, especially for the experiments carried out with low ambient
turbulence. However, it is found that both models heavily rely on numerical parameters and additional
validation is required to increase the confidence in their tuning. The data from the LES simulations
exhibited a clear wake centreline deflection for non-yawed rotor, an effect which is caused by the
rotating wake and the impinging non-uniform wind flow. It was shown in this study that when the effect
of this deflection component is not addressed, both the EPFL and the Jimenez models do not predict
well the deflections downstream of the turbine. After the non-yaw deflection component is linearly
added to the overall deflection obtained from the simulation, it was noticed how both the EPFL model
and the Jimenez model (using a kd parameter of 0.15) were able to fit reasonably well with the obtained
deflections.

6. Acknowledgements
This project has received funding from the European Union’s Horizon 2020 research and
innovation programme under grant agreement no. 727477 (CL-Windcon, website: www.clwindcon.eu).

Some of the data used for this work are part of UPWIND, a project funded by the European
Commission under the 6th (EC) RTD Framework Programme (2002- 2006) within the framework of
the specific research and technological development programme “Integrating and strengthening the
European Research Area”.
The authors thank the National Renewable Energy Laboratory (NREL) and in particular Paul Fleming
for kindly providing results from LES simulations of utility-scale prototype wind turbines.

References
[1] Lissaman P B S 1979 Energy Effectiveness of arbitrary arrays of wind turbines, Journal of
Energy, 3(6):323–8
[2] Jensen A 1983 Note on wind generator interaction, Risø DTU
[3] Ainslie J F 1985 Development of an eddy viscosity model for wind turbine wakes, Proc. 7th
British Wind Energy Association conference, Oxford
[4] Ainslie J F 1988 Calculating the flow field in the wake of wind turbines, Journal of Wind
Engineering and Industrial Aerodynamics, 27(1–3):213–24
[5] Bastankhah M, Porté-Agel F 2016 Experimental and theoretical study of wind turbine wakes in
yawed conditions, J. Fluid Mech., 806, pp 506-541
[6] Niayifar A, Porté-Agel F 2015 A new analytical model for wind farm power prediction, J. of
Physics, 625, 012039
[7] Ishihara T, Qian G W 2018 A new Gaussian-based analytical wake model for wind turbines
considering ambient turbulence intensities and thrust coefficient effects, Journal of Wind
Engineering and Industrial Aerodynamics, 177, 275–292
[8] Gebraad P M O, et al. 2016 Wind plant power optimization through yaw control using a
parametric model for wake effects a CFD simulation study, Wind Energy, 19:95–114
[9] NREL, FLORIS software, http://github.com/wisdem/floris
[10] TU Delft, FLORISSE software, https://github.com/TUDelft-
DataDrivenControl/FLORISSE_M103
[11] Sanders B, et al. 2011 Review of computational fluid dynamics for wind turbine wake
aerodynamics, Wind Energy, 14(7):799–819
[12] Churchfield M, et al. 2014 Overview of the simulator for wind farm application (SOWFA),
https://nwtc.nrel.gov/system/files/SOWFA_tutorial_05-20-2014.pdf
[13] van der Laan M, Sørensen N, et al. 2015 An improved k-ϵ model applied to a wind turbine
wake in atmospheric turbulence, Wind Energy, 18(5):889–907

19
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

[14] Ott S, Nielsen M 2014 Developments of the offshore wind turbine wake model Fuga, DTU Wind
Energy E, No. 0046
[15] Ozdemir H, Versteeg M C, Brand A J 2013 Improvements in ECN wake model, ICOWES2013
Conference
[16] CL-Windcon project 2018 Deliverable D1.4, Classification of control-oriented models for wind
farm control applications, http://www.clwindcon.eu/
[17] CL-Windcon website http://www.clwindcon.eu/
[18] CL-Windcon project 2018 Deliverable D3.1, Definition of wind tunnel testing conditions,
http://www.clwindcon.eu/
[19] CL-Windcon project 2018 Deliverable D3.4, Testing in the wind tunnel of wind turbine
controllers, http://www.clwindcon.eu/
[20] Svensson G, et al. 2011 Evaluation of the diurnal cycle in the atmospheric boundary layer over
land as represented by a variety of single-column models: The second GABLS experiment,
Boundary-Layer Meteorol, 140(2):177206
[21] Schepers J G, et al 2011 Analysis of wake measurements from the ECN Wind Turbine Test Site
Wieringermeer, EWTW, Wind Energy, 15(4):575–91
[22] Barthelmie R J, et al. 2011 Flow and wakes in large wind farms. Final report for UpWind WP8,
Risø DTU, 1765(EN)
[23] Frandsen S, et al. 2007 The shadow effect of large wind farms: measurements, data analysis and
modelling, Risø DTU, 1615(EN)
[24] DNV GL 2014, Windfarmer Validation report version 5.3, http://www.dnvgl.com/
[25] Hersbach H, et al. 2018 Operational global reanalysis: progress, future directions and synergies
with NWP, ECMWF
[26] Dyer A J 1974 A review of flux-profile relationships, Boundary-Layer Meteorology, 7(3):363
[27] Högström U 1988 Non-dimensional wind and temperature profiles in the atmospheric surface
layer: a re-evaluation, Boundary-Layer Meteorology, 42(1–2):55–78
[28] Peña A, et al. 2016 On the application of the Jensen wake model using a turbulence-dependent
wake decay coefficient: the Sexbierum case, Wind Energy, 19(4):763–76
[29] Holtslag A A M 1984 Estimates of diabatic wind speed profiles from near-surface weather
observations, Boundary-Layer Meteorology, 29(3):225–50
[30] Gryning S E, et al. 2007 On the extension of the wind profile over homogeneous terrain beyond
the surface boundary layer, Boundary-Layer Meteorology, 124, p. 251-268
[31] Pasquill F 1974 Atmospheric diffusion 2nd ed., Halsted Press
[32] Hansen K S, Larsen G C, Ott S 2014 Dependence of offshore wind turbine fatigue loads on
atmospheric stratification, Journal of Physics: Conference Series, 524:012165
[33] Mohan M 1998 Analysis of various schemes for the estimation of atmospheric stability
classification, Atmospheric Environment, 32(21):3775–81
[34] Quarton D, Ainslie J 1990 Turbulence in wind turbine wakes, J. Wind Eng., 14(1):15-23
[35] Crespo A, Hernandez J 1996 Turbulence characteristics in wind-turbine wakes, Journal of Wind
Engineering and Industrial Aerodynamics, 61(1):71–85
[36] Gunn K, et al. 2016 Limitations to the validity of single wake superposition in wind farm yield
assessment, Journal of Physics: Conference Series,749:012003
[37] Machefaux E, Larsen G C, Leon J P M 2015 Engineering models for merging wakes in wind farm
optimization applications, Journal of Physics: Conference Series, 625:012037
[38] Schlez W, Neubert A 2009 New developments in large wind farm modelling, Garrad Hassan
Deutschland GmbH
[39] Burton T, Jenkins N, Sharpe D, Bossanyi E 2011 Wind Energy Handbook, John Wiley & Sons
[40] Coleman R P, et al. 1945 Evaluation of the induced velocity field of an idealised helicopter rotor,
NACA, L5E10
[41] Jimenez A, Crespo A, Migoya E 2010 Application of a LES technique to characterize the wake
deflection of a wind turbine in yaw, Wind Energy, 13:559-572

20
WindEurope IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1222 (2019) 012004 doi:10.1088/1742-6596/1222/1/012004

[42] Martínez-Tossas L A, Annoni J, Fleming P A, Churchfield M J 2018 The aerodynamics of the

curled wake: a simplified model in view of flow control, Wind Energy Science Discussions,
7:1–17
[43] Fleming P, et al. 2018 A simulation study demonstrating the importance of large-scale trailing
vortices in wake steering, Wind Energy Science, 3(1):243–55
[44] Zahle F, Sørensen N. 2008 Overset Grid Flow Simulation on a Modern Wind Turbine, 26th AIAA
Applied Aerodynamics Conference, Honolulu, Hawaii
[45] Sakagami Y, Santos P A A, et al. 2017 Low turbulence intensity and low wind shear effects on
annual energy production over the northeast coast of Brazil, Wind Europe Conference

21