DTU Wind Energy E 0046

Downloaded from orbit.dtu.
dk on: Sep 16, 2020
Developments of the offshore wind turbine wake model Fuga
Ott, Søren; Nielsen, Morten
Publication date:
2014
Document Version
Publisher's PDF, also known as Version of record
Link back to DTU Orbit
Citation (APA):
Ott, S., & Nielsen, M. (2014). Developments of the offshore wind turbine wake model Fuga. DTU Wind Energy.
DTU Wind Energy E, No. 0046
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright
owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
 Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
 You may not further distribute the material or use it for any profit-making activity or commercial gain
 You may freely distribute the URL identifying the publication in the public portal
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
Developments of the offshore
wind turbine wake model Fuga
Department of
E Report 2014
Wind Energy
Søren Ott and Morten Nielsen
DTU Wind Energy E-0046
January 2014
Abstract This is the final report of the project entitled Risø DTU Modelling Services
carried out by DTU Wind Energy (formerly known as Risø National Laboratory) as part
of the Carbon Trust’s Offshore Wind Accelerator Stage 2 under a contract with Carbon
Trust. The project is a follow–up to a Carbon Trust’s Offshore Wind Accelerator Stage 1
project called Linearized CFD Wake models. The earlier project resulted in the development,
implementation and validation of the Fuga model. Fuga is a linearized CFD model that can
predict wake effects for offshore wind farms. The main purpose of Stage 2 is to add more
features to Fuga and turn it into a useful tool for offshore wind farm developers. The new
features consist in
• Flexibility. Including the ability to cope with several types of turbines in the same
project, thus making it possible to predict inter farm interactions. The graphical user
interface has been greatly improved and a number of input/output facilities have been
added.
• Stability effects. The effect of stability has been added through a modification of the
eddy viscosity based on Monin–Obukhov theory. The numerical solver developed in
Stage 1 has been generalized in order to make it deal with the modified equations.
• Meandering. Meandering has been included in the form of a post processing of the
model results that bend and twist the wake centreline. The meandering centrelines are
calculated using a Gaussian process developed on the basis of measured spectra. An
analysis of meteorological data from Horns Rev has been made in order to quantify the
impact of non–stationarity of the wind direction. The results are generalized so as to
account for the uncertainties imposed by a ten minute mean value trend as well as by
the distance between turbines and the met mast.
The old model has been validated against a number of data sets. Some of these tests have
been repeated in order to demonstrate and validate the new model features. Production data
from Horns Rev 1 have been re–analysed using well defined selection criteria for which the
developed uncertainty models apply, and a comparison with data is made. Even if the model
predictions fall within estimated error bars, the model seems to over predict the measured
efficiencies by a few percent.
The model works best for unstable, neutral and light stable conditions whereas the results
for stable and very stable conditions are questionable. We suspect this is caused by a failure
of the numerical solver that becomes progressively more severe as the stability increases.
Approved by: Hans E. Jørgensen
Checked by: Morten Nielsen
In no event will the Technical University of Denmark (DTU) be liable for any damage,
including any lost profits, lost savings, or other incidental or consequential damages arising
out of the use or inability to use the results presented in this report, even if DTU has been
advised of the possibility of such damage, or for any claim by any other party.
Contents
1 Introduction 5
2 Flexibility 6
3 Atmospheric stability modelling 7

3.1 Theory 7
3.2 Model equations 10
3.3 Numerical methods 11
4 Wake meandering 12
4.1 Data interpretation 12
4.2 Turbulent scales 16
4.3 Wake meandering modelling 19
5 Rotor averaging 23
6 Validation 25
6.1 Repeats of old tests 25
6.2 New Horns Rev 1 test 27
7 Guidelines 32
7.1 Stability input to Fuga 32
7.2 Roter averaging 32
7.3 How to make a climatology with stability 32
7.4 Fuga user interface 35
References 41
8 Appendix A: The Fuga help system 42
DTU Wind Energy E-0046 3

1 Introduction
This is the final report of the project entitled Risø DTU Modelling Services carried out by
DTU Wind Energy (formerly known as Risø National Laboratory) as part of the Carbon
Trust’s Offshore Wind Accelerator Stage 2 under a contract with Carbon Trust. The project
is a follow–up to a Carbon Trust’s Offshore Wind Accelerator Stage 1 project called Lin-
earized CFD Wake models. The main outcome of this earlier project was the Fuga wake
model, which is based on a linearized CFD model using the ’simple’ closure. The model
is described in Ott (2011), the final report of the first stage of the project.1 . The present
project aims at extending and refining Fuga and turn it into useful tool for the quantification
of wake effects in wind farms and between wind farms.
The report constitutes Deliverables 3 and 9 under Task 5 (cfr. Trust 2011). There are
thus three purposes:
1. Technical documentation of added model features
2. Evaluation against measurements
3. User guidelines.
The main objective of the present project is to make the Fuga model a more useful tool
by adding some new features. The objectives fall into three categories:
• Flexibility.
• Stability effects.
• Meandering.
Before describing the developments in more detail we briefly recapitulate the status of the
’old’ model. The ’old’ Fuga model (i.e. version 1.4) was validated as part of the Linearized
CFD project in Carbon Trust’s Offshore Wind Accelerator Stage 1, cfr. Ott (2011) and Ott
et al. (2011). Actually three different versions of the model based on three different closures
were evaluated. For each closure model predictions were compared to production data from
Horns Rev 1, Nysted and Nibe wind farms. Gribben, de Villiers, Hui, Balcombe, Harrington,
Peace, Mallins, Housley, Kristensen and Bech (2011) made a similar analysis of data from
the North Hoyle wind farm. Furthermore, results from the ’simple’ version of the model were
compared with wake deficits measurements taken at met masts located 2 and 6 km in the
lee of Horns Rev 1. The following points summarize the results of the model:
• The ’simple’ closure was clearly better than the two other, more elaborate, and thus
supposedly better, closures.
• Rather large discrepancies were found for data for narrow wind direction bins. This was
partly explained by inaccurate wind direction data. The wind vane was out of order
and one of the turbines was used as anemometer (inferring the wind direction from the
yaw angle). Meandering was the second, more fundamental explanation.
• When data were collected in large enough wind direction bins, the effect of data being
’misplaced’ in the wrong bin becomes negligible, and the comparison becomes more
benign towards the model. For relatively wide bins the comparison with data strongly
favors the ’simple’ closure. In fact this version of the model leaves very little room for
improvements.
• Model predictions (using the simple closure) of wake deficits show excellent agreement
with data measured at the met mast M6, located 2km behind the last row, and M7,
located 7km behind the last row.
It should be noted that we have attempted to make the simplest possible (and still mean-
ingful) adjustments of the model.
1A publicly available version is found in Ott, Berg and Nielsen (2011)

2 Flexibility
A certain lack of flexibility characterizes the old model. Most noticeable is the fact that it
only works for one type of turbines at a time. This excludes calculations of farm–to–farm
interactions unless they are assumed to be equipped with the same type of turbine. The new
model is capable of working with WAsP project files. Thus WAsP is used to define turbine
types and locations and the workspace generated by WAsP is used as input to Fuga. One
implication of this is that the turbines can be split into groups (and even sub–group) each
with their own turbine type. Another implication is that a different wind climate can be
assigned to the individual turbines or to turbine groups. In this way the model can take
spatial variations of the wind climate into account. Fuga handles this by making a separate
calculation for each wind climate to produce the results the turbines associated with it. Thus
Fuga still only uses one wind climate at a time to calculate wake effects. This is permissible
because the largest wake effects are due to neighbouring turbines.
The old model only used the velocity at the hub height plane to estimate the thrust
assuming that the trust scales with U 2 (zhub ), the ’free stream’ velocity at hub height. With
several turbine types the flow field has to be calculated at several different heights, and the
availability of the whole 3D flow field can be used also to make a (supposedly) more accurate
estimate of the thrust by replacing U 2 (zhub ) with a numerical averaging of U 2 over the rotor
plane. These changes have resulted in a major revision of data structure and functionality
of the graphical user interface.
Numerous changes have been made to expand and improve the functionality of the graph-
ical user interface (GUI). The built in help system documenting the changes that were made
is included as Appendix A.
6 DTU Wind Energy E-0046

3 Atmospheric stability modelling
The old model did not incorporate effects of stability. In the verification data were not split
up according to stability, hence the comparison could be misleading, and it would perhaps
have been better to pick only near neutral data. However, the inclusion of stable and unstable
data may not actually make much difference in terms of predictions of net annual energy
production. Long range, farm–to–farm interactions could, on the other hand, depend more
strongly on stability. The new model is based on the ’simple’ closure, but with a stability
dependent eddy viscosity as explained below.
3.1 Theory
In the new model atmospheric stability is represented as a change of the eddy viscosity
profile of the approach flow. The eddy viscosity K determines the velocity profile through
the relation
dU
K(z) = u2∗ (1)
dz
Following the ’simple’ philosophy we assume that the eddy viscosity is unaffected by the
presence of a wake. Measurements in wakes show increased levels of turbulence, but the
length scale is decreased at the same time so that the effect on K seems to be limited.
The buoyancy term in the momentum equation is ignored. This eliminates the need for
a (potential) temperature equation. A temperature equation coupled with the momentum
equation via a buoyancy term would produce solutions that exhibit gravity waves. It is likely
that gravity waves can have interesting effects under very stable conditions, e.g. when the
turbines excite standing waves in the farm. This has been observed in the lee of islands with
a small mountains on them, where it causes the wind to speed up instead of slow down. It
would not have been very difficult to include such buoyancy effects, but since they are not
likely to be of much importance, except perhaps for some very rare cases, we choose to keep
things as simple as possible and neglect the coupling via buoyancy. Buoyancy is therefore
only modelled indirectly through the eddy viscosity K. In the old model the eddy viscosity,
K, was set equal to κu∗ z, which is its value in a flat, homogeneous terrain (such as an open
sea) under neutral, barotropic conditions. For non–neutral conditions we will use the eddy
viscosity dictated by Monin–Obukhov (M–O) similiarity theory, Monin and Obukhov (1954)
(see also Monin and Yaglom 1975).
M–O theory introduces the Monin–Obukhov length scale L defined as
T u3∗ T u2∗
L=− = (2)
κghwθi κgθ∗
p
where T is the absolute temperature at the surface, u∗ = −hw0 u0 i is the friction velocity,
g is the acceleration of gravity and θ is the potential temperature and θ∗ = hwθi/u∗ . L/u∗
is a corresponding time scale and θ∗ is a temperature scale. When these scales are used to
write the equations in non–dimensional form a small miracle happens: the equations are the
same no matter what L is! It should be noted that ’the equations’ are the Navier-Stokes
equations in the Boussinesq buoyancy approximation and without the Coriolis force. The
use of the eddy viscosity concept is sometimes also called the Bousinnesq approximation,
but that assumption is not necessary to derive the theory, nor do we need to postulate any
closure. We are, in other words, talking about equations that have fully realistic turbulent
solutions. In flat terrain it gets even better because the boundary conditions for the mean
fields become independent of L too. As a result the mean velocity profile must have the form
U (z) = u∗ (f (z/L) − f (z0 /L)) (3)
where f is a universal function that can be determined from measurements. It turns out

that f (ζ) has logarithmic pole at ζ = 0 so that it can be written as
1
(log ζ − ψm (ζ))
f (ζ) = (4)
κ
where ψm is more well behaved. The constant in front is the inverse von Karman constant.
Using ψ (3) becomes
u∗
U (z) = (log z/z0 − ψ(z/L) + ψ(z0 /L)) (5)
κ
This is Monin–Obukhov similiarity theory in a nutshell.
If we assume that there is such a thing as an eddy viscosity then (1) yields
u∗ κz
K= (6)
φm (z/L)
where
0
φm (ζ) = 1 − ζ ψm (ζ) (7)
The mean potential temperature profile has a similar form2
θ∗
hθ(z)i = T + (log z/z0 − ψh (z/L) + ψh (z0 /L)) (8)
κθ
There is a simple relation between temperature and potential temperature:
θ(z) = T (z) + g/Cp z (9)
where T (z) is the absolute temperature at height z and Cp is the specific heat of air at
constant pressure.
In M–O theory stability varies with height. Near the ground the profiles are dominated
by logarithm and close to neutral while the ψ functions become more an more important
as the non–dimensional height ζ ≡ z/L increases. In this respect M–O theory differs from
the once very popular Pasquill (1974) stability class system, where stability is independent
of height. For wind turbines zhub /L would be a relevant stability measure, but since the
hub height does no vary very much for offshore turbines we follow Gryning, Batchvarova,
Brummer, Jørgensen and Larsen (2007) and use 1/L to define stability classes (see table 2
in this report).
In the new mode the look–up tables have three entries: kz0 , β and ζ0 ≡ z0 /L. The old
model had only two (kz0 and β, see Ott et al. 2011). The extra dimension of the tables
make them considerably more voluminous.
M–O theory assumes equilibrium with constant fluxes that do not vary with height. For
offshore wind farms there could be a problem because non–neutral situations often corre-
sponds to situations that are created by a temperature difference between sea and land.
If the wind blows from land to sea and the land is warmer than the sea then the sea will
cool the air and create a stable layer with a capping inversion that grows very slowly with
the distance from shore (the fetch). While the stable layer cools it approaches a neutral
layer, and the assumption of equilibrium could be violated. Lange, Larsen, Højstrup and
Barthelmie (2004) argue that this seems to be the case for the Nysted wind farm. Cold air
on a warm sea will of course create an unstable layer that also decays to a neutral layer. The
stability over open sea thus tends to be neutral, although sea temperature differences over
long distances can make stability deviate from neutral and radiative cooling from clouds
can create convection. Sathe, Gryning and Peña (2011) analysed met data from Horns Rev
and found that unstable conditions are much more common for NW winds than for SW
and explains it as caused by the general N–S gradient of the sea temperature. Stability
generated in this way should be similar to stability over a land surface, whereas stability
generated by temperature differences between land and sea is probably more complex. A
further investigation of this is needed, but this falls outside the scope of the present project.
2 ψ (z /L) is usually small and the term is often neglected.
h 0

Theory says that ψm and ψh are universal functions of ζ in the equilibrium surface layer
under perfect, barotropic conditions. In practice these requirements are only met approxi-
mately and plots of ψ = U κ/u∗ − log(z/z0 ) against z/L always a great deal of scatter. There
are therefore many different bids on what the ψ functions should look like. It is conventional
to use the functional form
φ(ζ) = (1 + a ζ)n (10)

with two different sets of constants a and n for ζ > 0 (stable) and ζ < 0 (unstable). With
two ψ functions we end up with 4 sets of constants. We have chosen the following functions
which are consistent with the data analysis of Högström (1988):

(1 + 5.0 ζ) for ζ > 0
φm (ζ) = (11)
(1 − 19.3 ζ)−0.25 for ζ < 0

(1 + 7.8 ζ) for ζ > 0
φh (ζ) = (12)
(1 − 12 ζ)−0.5 for ζ < 0
The uncertainty of these profiles is considerable, especially for strong stability where not
much is known for values of z/L larger than about 1. We are therefore in trouble when
zhub > L > 0. Very stable conditions with L less than about 100m are therefore out of
Fuga’s reach. It should be noted that the boundary layer depth generally decreases with L
and for extreme stability the boundary layer may not even reach up to the turbine rotors,
and the surface layer will be even more shallow. In such situations the shape of the wind
profile will depend on many factors such as the sea–land effects mentioned above, effects of
baroclinity, the presence of a low level jet etc. It therefore becomes more and more impossible
to maintain the assumption of a universal and predictable behaviour as stability increases.
The fluxes of momentum and heat that go into the definition of L can be measured
directly with a sonic anemometer. In case sonic anemometer data are not available L can
be estimated from wind and temperature profiles. Over a sea surface stability is driven by
the air to water temperature difference. It is therefore important to have measurements
of temperature in the water as well as in the air. In principle two air temperatures taken
at different heights could also work, but the difference is so small that it easily becomes
very inaccurate. It is therefore essential that offshore met masts include water temperature
measurements. L can be estimated from ten minutes values of air and water temperatures,
a wind speed, current and an estimate of the surface roughness. We will assume that the
current is negligible and obtain the surface roughness from Charnock’s (1955) relation:
z0 = Ac u2∗ /g + Bc ν u∗ (13)
where Ac is the Charnock constant (following Peña and Gryning (2008) we use Ac = 0.012),
ν is the molecular viscosity, and Bc = 0.12, Grachev and Fairall (1996). A second term
has been added in order to accounts for the limit of laminar flow at very low wind speeds.
Using M–O profiles we then have enough equations to determine L. In general the equations
must be solved numerically, but for near neutral conditions there are simplifications. From
observations of wind speed U taken at height zu , air temperature ta taken at height zt and
sea surface temperature ts we can form a so–called bulk Richardson number Rb :
gzref (θ(zt ) − Ts )
Rb = (14)
Ts U 2
where zref is a reference height of our choice - its only purpose is to make Rb non–dimensional.
Inserting profile functions into (48) we find the following relation
(log zu /z0 − ψm (zu /L))2
zref /L = Rb (15)
log zt /z0 − ψh (zt /L)
For near neutral conditions we have zt /L ∼ zu /L ∼ 0 so that we are allowing us to drop ψm
and ψh . Using a good guess for the surface roughness, say z0 = 0.1 mm, we then arrive at

the estimate
1 Rb (log zu /z0 )2
≈ (16)
L zref log zt /z0
This relation does not require any knowledge of the universal functions. Evaluating Rb from
data is straight forward and (16) gives good estimates for |z/L| < 0.1. Thus equation (16)
suggests an approximate relation between zref /L and Rb even if an exact relation does not
exist within M–O theory. Grachev and Fairall (1996) derived such an empirical relation,
which is widely used. However, it is only valid for the special case zref = zu = zt = 10meter,
and it is not obvious how to generalize it to other heights. Consequently many different
methods are in use. In order to put an end to this confusion we have made a tool called
AMOK that solves the equations numerically and yields L from ts , ta and U measures at
any heights.
The numerical solution starts with (16) to obtain a first guess on 1/L. This is then
used to evaluate ψm and ψh so that U and θ can be evaluated from (5) and (8) and z0 is
calculated from (13). The next iteration is the same except that it starts with (48) rather
than 16. The scheme is very robust and seems always to converge to a solution provided one
exists. It converges rather slowly, however, and an alternative scheme is used to converge
down to machine precision. The second scheme is a 2D Newton-Ralphson iteration for the
two variables U and θ with z0 eliminated by means of (13). The combined scheme is fairly
robust and efficient except for rare cases with a combinations of extremely stable conditions,
low wind speed and large difference of zu and zt . In the data studied so far cases with no
solution are extremely rare, and only occur below the cut in speed for most wind turbines.
3.2 Model equations

The governing equations are basically the same as was used in Ott et al. (2011) for the
’simple’ closure. They are formulated in a mixed–spectral setting, where the wind field
(u, v, w) and the pressure p have been Fourier transformed along the two horizontal directions
x and y. The fields are therefore functions of two wave numbers k1 k2 and z. The equations
read
∂u0 ∂ ∂u1 ∂K
u0 ik1 u1 + w1 = −k2 K u1 + K + ik1 w1 − ik1 p1 + f11 (17)
∂z ∂z ∂z ∂z
∂ ∂v 1 ∂K
u0 ik1 v 1 = −k2 K v 1 + K + ik2 w1 − ik2 p1 (18)
∂z ∂z ∂z
∂ ∂w1 ∂K ∂w1 ∂p1
u0 ik1 w1 = −k2 K w1 + K + − (19)
∂z ∂z ∂z ∂z ∂z
∂w1
ik1 u1 + ik2 v 1 + = 0 (20)
∂z
where k2 = k12 +k22 . The difference is that K now depends on stability as described in section
3.1. The boundary conditions are:
u(k1 , k2 , z0 ) = 0
v(k1 , k2 , z0 ) = 0
w(k1 , k2 , z0 ) = 0
u(k1 , k2 , zi ) = u0 (zi )
v(k1 , k2 , zi ) = 0
w(k1 , k2 , zi ) = 0 (21)

3.3 Numerical methods
The friction velocity is eliminated from the equations by scaling the velocity components
u∗ , and pressure and rotor forcing with u2∗ as described in Ott (2011). In the mixed–spectral
formulation there is only one independent variable, z, since k1 = k cos β and k2 = k sin β
merely act as constants. Using the variable kz instead of z eliminates k, but we still have β,
and kz0 from the boundary conditions (and from u0 ). We also have ζ which we write as
z kz(z0 /L) kζ0
ζ= = = (22)
L kz0 kz0
where we have defined
z0
ζ0 = (23)
L
In the old model the look–up tables had two entries: kz0 and β. Now we have three: kz0 , β
and ζ0
The solution method described in Ott et al. (2011) was applied with a modification that
takes the changed definition of K into account. In the original procedure the variable z was
replaced with the variable t = log z/z0 in the lowest part up to kz = 1, while s = kz was
used in the upper layer. In the new model we generalize the definition to
t = log z/z0 − ψ(z/L) + ψ(z0 /L) = log s/kz0 − ψ(s/kz0 ζ0 ) + ψ(ζ0 ) (24)
We are, in other words, essentially using the unperturbed velocity to indicate the height
above the ground. The second identity show that t can be expressed in terms of s and the
two parameters kz0 and ζ0 . The following identity can be used to change the equations to
the new variable
∂ K ∂
= (25)
∂t κu∗ ∂z
The numerical solution to the linearized equations is the Achilles heel of the model. It
was only with great difficulty we managed to solve the equations for the neutral case, and
from start there was no guarantee that it would work for all stabilities. The problem is
that the equations have particular solutions that increase or decrease extremely fast, and
finding the well behaved linear combination that solves the problem is like finding a needle
in a haystack. The problem becomes less severe when K is large. This happens on the
unstable side which is therefore quite benign. On the stable side, however, K gets smaller
and smaller as the stability increases. It was therefore expected that the numerical solution
would break down at some point, but hopefully only at very strong and rare stabilities. The
first expectation came true, but unfortunately the solutions seem to misbehave even at quite
moderate stabilities. It was not the intention to develop new numerical solution methods in
the present project, but a few ideas were tried none of which lead to an improvement. This
problem is something that needs to be looked at in the future.

4 Wake meandering
4.1 Data interpretation
Before we proceed to the description of the wake meandering model, we need to discuss the
interpretation of meteorological data in terms of input to flow models.
Real world Predictions
4D fields
Weather
Model world
Measurements Idealized flow
Met mast Homogeneous
SCADA Stationary
The diagram illustrates the various ingredients that go into the validation of a wake
model. We have the real world with its complex fields (velocity, temperature etc.) and
complicated interactions. We don’t have full access to the real world, but have to be make
do with measurements such as met mast data and SCADA data. These are essentially point
measurements and we have to make some kind of interpretation in order to fill the gaps and
create an idealized model world which our wake model can work with. In the model world
the fields have to be statistically stationary because otherwise we cannot make a Reynolds
decomposition and therefore we cannot apply a RANS model. Horizontal homogeneity is
also a necessary assumption in our case. Once we have decided how to interpret the data
we can make model predictions which finally can compared to data. Strictly speaking, you
cannot be compare directly to the real world which remains a bit on the sideline in the
diagram. If the comparison is good there are two possibilities: either the data interpretation
and the model are both correct or they are both wrong. Two wrongs can make a right, and
this may if fact occur if the data interpretation is incorrect and the model is tuned to make a
good comparison. It is therefore necessary to discuss data interpretation. In order to do this
in a simple way we will make two different interpretations and work out the consequences
for the data comparison. We will also focus on the wind direction because wake effects are
very sensitive to this parameter. Similar considerations can be given to other parameters
such as wind speed and stability measures, but the wind direction has the largest impact.
The traditional interpretation of met data is to regard each ten minutes period as be-
longing to a statistically stationary time series with (abstract) mean values equal to the
corresponding ten minutes averages. In figure 1 this corresponds to the black, piecewise con-
stant curve. The implication is that a single mean value of the wind direction is assigned

290
285
q @degD
280
275
270
t @sD
600 1200 1800 2400 3000 3600 4200 4800 5400 6000
Figure 1. Green: wind direction time series. Black: mean value, traditional interpretation.
Red: mean value, alternative interpretation.
every ten minutes and used as input to the model. This of course does not make the average
wind direction equal to the ’true’ mean wind. We could shift the ten minutes periods by e
few minutes and get different mean values with the same data. We will therefore investigate
the implications of making a more realistic, alternative interpretation.
The alternative interpretation is illustrated by the red curve. Here the mean value is
changing continuously in time in a piecewise linear manner connecting the ten minutes
averages. This mean that each ten minutes period corresponds to a range of mean values
depending on the slopes of the linear pieces. Instead of feeding the model with one wind
direction for each period, we should feed the model with a range of values distributed around
the observed average value. In other words, we are still treating the model world flow as
statistically stationary, but allow more variation of the wind direction input than in the
traditional approach.
Figure 2 shows a histogram of measurements of ∆θ, the difference between consecutive
ten minutes averaged of the wind direction. The red dots show a fit to an exponential
distribution:
P∆θ (x) = β/2 e−β|x| (26)

Further data analysis has shown that the parameter β depends on both wind speed and
stability, and the following expression has been fitted;
a1 ζ0 + b1 u∗
β= (27)
1 + a2 ζ0 + b2 u∗
The values of the fitting constants are listed in Table 1. These values are default values in
Fuga. They could be used tentatively for other locations than Horns Rev, even if site specific
values based on local measurements are to be preferred.
Table 1. Constants to be used in (27).
Stability a1 [-] a2 [-] b1 [s/m] b2 [s/m]

Stable 2.51 103 -6.80 104 3.55 4.55
Unstable -3.95 104 -2.01 105 3.70 4.99
We need the distribution Pδθ of the variable δθ ≡ θ − θ where θ is the wind directions to
be used as model input and θ is the observed ten minutes average wind direction. In figure
1 δθ represents the difference between the red line and the black line. Each linear piece of
the red line corresponds to an interval of mean values with
−|∆θ|/2 < δθ < |∆θ|/2 (28)
Thus all the flow cases corresponding to the column around ∆θ in the histogram in figure 2
should be distributed evenly on this interval. In terms of probability this can be expressed
as

5000
4000
3000
f HDqL
2000
1000
0
-15 -10 -5 0 5 10 15
Dq
Figure 2. Histogram of ∆θ, the difference between successive ten minutes averaged of the
wind direction. Data were from met mast M7 located to the east of Horns Rev 1. The red
dots is a fit to an exponential distribution.

 P∆θ (∆θ) d∆θ for −|∆θ|/2 < δθ < |∆θ|/2
|∆θ| dPδθ (δθ) = (29)
0 otherwise

Integrating this we find

−2|x| ∞
P∆θ (x0 ) 0 P∆θ (x0 ) 0
Z Z
Pδθ (x) = dx + dx = β Γ(0, 2β|x|) (30)
−∞ |x0 | 2|x| x0
where we have used (26) and
Z ∞
Γ(k, x) = tk−1 e−t dt (31)
x
is the incomplete gamma function. The way to apply the alternative interpretation is to
convolute the distribution of observed average directions (which could be a rectangular
distribution on a bin) with Pδθ . This is best done in the Fourier domain, where it is just a
matter of multiplying with a filter φδθ given by
Z ∞
2β k
φδθ (k) = Pδθ (x) e−ikx dx = Arctan (32)
−∞ k 2β
The width of √the filter Pδθ , determined as the square root of the second central moment,
is equal to 1/(β 3).
Figure 3 shows results from a wake model test made in the EERA-DTOC project. The
graphs show normalized power deficits in the crosswind rows of Horns Rev 1 for Fuga and
7 other wake model. A narrow, 5 degree wide, wind direction bin centrered around a wind
direction that runs parallel with the turbine rows has been chosen. For the narrow bib all the
models predict larger deficits than was measured. This test is based on the conventional data
interpretation. The red curve is the Fuga results when we use the alternative representation.
For historical reasons a 5 degree wide square filter√was used instead of (30). The width√
based on the second central moment is therefore 5/ 12 degrees. Equating this to 1/( 3β)
we get β = 0.40 which is exactly what (27) assuming neutral conditions and U=8m/s.
The improvement of the comparison with data using the alternative interpretaion is quite
evidently. It is worth noting that the interpretation does nothing to the model itself, it
is only the way model results are binned before comparison with data which is affected.
The same kind of tests were made by Barthelmie, Hansen, Frandsen, Rathmann, Schepers,
Schlecz, Philips, Rados, Zervos, Politis and Chaviaropoulos (2009) in the UpWind project

where bins as narrow as ±1 degree were used. The intension is to test the ability of wake
models to handle multiple wakes (wakes in wakes), but it is not possible to determine the
cause of the wake deficits being larger than observed. It could be caused by the models wakes
being too narrow or too strong or it could be caused by the wind direction being uncertain.
The present analysis does not solve this dilemma, but it shows that the uncertainty of the
wind direction should not be neglected. Using this kind of test for model tuning is therefore
highly problematic.
Data issues and problems with establishing flow cases has been discussed before. Barthelmie
et al. (2009) list a number of issues, but largely see it as problems that the models should
be able to deal with. Here we see it more as a measurement issue, which can be dealt with
by further analysis of met data, preferably from the site in question.
Fuga + 5deg filter
Technical University of Denmark
Figure 3. A slide showing wake model results averaged over a narrow wind direction bin.
The stationarity analysis can be extended to include effect of inhomogeneity in a simple

way. The wind direction measured on a met mast is likely to differ from the wind direction
at a turbine placed far away from it. If the met mast is located directly upwind with respect
to the turbine, there will be a delay X/U before the air that pass by the mast will pass by
the turbine. By Taylor’s hypothesis we assume that the wind field is frozen and just slides
downwind with velocity U without changing. This assumption is of course not perfect, but
it is correct in an average sense. Vincent, Larsén, Larsen and Srensen (2013) investigated
the correlation between wind speed measurements taken at a pair of offshore met masts and
the spatial and temporal decay of the correlation. This study confirms that the signals are
best correlated when a time delay equal to the travel time X/U is applied. It also shows that
only the low frequency parts of the signals are correlated and that the decay the Fourier
component corresponding to frequency f goes like e−af |X|/U , where a is a constant that
depends on the angle between the wind direction and the line connecting the two masts (or
mast and turbine). In the downwind case a = 4.1 while a = 7.7 in the crosswind case. Travel

times of several minutes are common in offshore wind farms and hence eddies corresponding
to time scales 4 to 8 times longer than the travel time will be correlated. These eddies are
found in the second group marked ’drift’ in figure 6. We therefore assume that the trend
in the ten minutes average wind direction persists during the travel from the mast to the
turbine. In other words, we simply extrapolate the red segment in figure 1 by the delay
in order to predict the most likely behaviour at the turbine. The corresponding segment is
illustrated as the blue segment in figure 4. Finally we assume that this procedure can be
followed even in situations where the turbine is not directly downwind of the mast. It is a
crude model, but it should give some of the right tendencies.
Figure 4. The blue line corresponds to the assumed wind direction interval seen at distance
X from the mast.
In terms of math we simply replace (29) by
t 1 1 t

 P∆θ (∆θ) d∆θ for T − 2 |∆θ| < δθ < |∆θ| 2 + T
|∆θ| dPδθ (δθ) = (33)
0 otherwise

where t = X/U and T is the averaging time (e.g. ten minutes). Inserting (26) into (33) and
integrating yields
 n o
β 2β|x| 2β|x|

 2
 Γ(0, 2t/T +1 ) + Γ(0, 1−2t/T ) for t < T /2
Pδθ (x) = n o (34)
 β Γ(0, 2β|x| ) − Γ(0, 2β|x| )

for t > T /2

2 2t/T +1 2t/T −1
Again the Fourier transform is relatively simple:

Z ∞
β k(1 + 2t/T ) k(1 − 2t/T )
φδθ (k) = Pδθ (x) e−ikx dx = Arctan + Arctan (35)
−∞ k 2β 2β
Figure 5 shows how a 5 degree wide bin with observed average wind directions at M7
translates to a wider distribution of observations at M6. The red line is the model, the
convolution of (34) with a 5 degree wide rectangular filter. The observed distribution is a
bit wider than the model distribution, but it is not completely far out. This could perhaps
be expected since we did not take decay of eddies into account.
4.2 Turbulent scales

Before we describe the meandering model we need to elaborate on the various time and length
scales that are involved and explain in more detail the assumptions that have been made
about the physical processes that govern the wake behaviour. Figure 6 shows an old spectrum
taken from Courtney and Troen (1990), where we have identified five different regimes. The
spectrum is made from one full year of wind data and thus covers a wide range of time
scales. Note that areas below the graph are proportional to the corresponding turbulent

Figure 5. Green line: pdf of average wind directions observed at M7 with bin centre shifted to
zero. Dots: pdf of the corresponding average wind direction observed at M6. Red line: model
prediction of the pdf at M7.
kinetic energy. The high peak at around 10−6 Hz is due to diurnal variations and is part of
a big ’mountain’ of low frequencies. Although these slow fluctuations represent almost all
the kinetic energy, they effectively just determine the mean wind speed and direction in any
given ten minutes period, and we would not normally refer to this as ’turbulence’. At slightly
higher frequencies we have the fluctuations that can make the wind direction change over
10 minutes, which were discussed and analysed in section 4.1. These variations are caused
mainly by mesoscale mechanisms such as the passage of weather systems, convection cells,
gravity waves or clouds. Near the surface this mainly affects horizontal motion are mainly
horizontal even if the flow as such is not two dimensional. Lindborg (1999) analysed long
range spatial velocity correlation data derived from airplane flights and saw two ranges. For
low wave numbers k < 2π/500km) the spectrum varies as k −3 which is due to an enstrophy
cascade that represents true 2D behaviour. For larger k in the range 2π/30km–2π/500km
the spectrum varies as k −5/3 , which could in principle be caused by a 2D energy cascade,
but Lindborg argues against this and prefers to interpret it as 3D behaviour caused e.g. by
gravity waves which can also produce such a range. When in figure 6 the local maximum
at ∼ 10−2 Hz is taken to divide ’2D’ from ’3D’ behaviour, the ’2D’ should be taken with a
grain of salt. Even if the motion is mainly horizontal the flow a mixture between 2D and 3D
behaviour. The next blob marked ’meandering’ represents motion that takes place ’on top
of’ the drifting mean wind. In figure 1 it is contained in the excursion of the green curve
from the red curves. These scales are responsible for the distortions of the wake centrelines
that are referred to as meandering. We could characterize the ’meandering’ eddies as large
enough to move a cross section of the wake as a whole, yet small enough the make the
centreline wiggly–waggly. Still smaller scales (larger frequencies) represent 3D eddies that
are comparable in size with the wake cross section but too small to contain it. This regime is
marked ’scrambling’ because it represents eddies that make a lot of mess by moving around
parts of the wake. Modelling the effects of this is, to say the least, not straight forward, and
today it can only really be done with large eddy simulations. Finally, we have the eddies
that are much smaller than the wake cross section. They only stir the wake locally and cause
momentum to spread in diffusive manner. In this way they tend to smooth out the mess
created by the slightly larger eddies, but they never really succeed in doing so. It is only the
effect of the smallest eddies that can truly be represented by an eddy viscosity.
The main features of the spectrum shown in figure 6 are common to datasets from mid
latitudes. The Australian dataset studied by Ayotte, Davy and Coppin (2001) shows similar
trends. In general the high frequency, 3D part is more site specific than the low frequency
2D part. The 3D turbulence is mainly created by surface friction and the turbulent kinetic
energy is proportional to u2∗ . The level therefore depends on the the wind speed (at 10m, say),
but also on terrain since u∗ increases with increasing roughness (for a fixed U ). Therefore

gimes
2D
3D
en 2011
Courtney & Troen 1990

Denmark
Figure 6. Spectral regimes - see text for an explanation. The spectrum is from Courtney &
Troen (1990).
there is generally more turbulence in this part of the spectrum for onshore sites than for
offshore sites. Larsén, Vincent and Larsen (2012) found that the low frequency mesoscale
part, is very similar from site to site, even from onshore to offshore sites. Moreover, the low
frequency part of the spectrum does not seem to depend on stability nor on the average wind
speed. This makes sense if we think of phenomena such gravity waves, clouds or convection
cells which can be transported around by the main wind without being dynamically coupled
to it.
Eddy viscosity can only truly represent the smallest eddies, but for a realistic RANS
model the eddy viscosity must represent a larger part of the turbulence. It could be argued
that the result of the action of ’scrambling’ eddies work a bit like viscose diffusion when a
time averaging is applied. As we go to larger scales, the eddy viscosity becomes less and less
realistic and we can ask ourselves whether we should consider the larger scales as described
by the model. If we go to the limit and include all scales, we end up with a model describing
something like one year averaged wakes. Depending on the wind rose, the wakes would
spread out from the turbines in all directions and the model would not be able to handle
shadow effects at all. We are therefore forced to exclude too large eddies from the model
world. This means that we can only predict what would happen if the large eddies were not
there. Comparison with data from the real world then becomes complicated by the fact that
large eddies do exist in the real world. It is also difficult to say exactly where we should draw
the line between ’large’ and ’small’ eddies. We could perhaps answer such questions if we

had a stringent derivation of the model equations from first principles, and if we had a deep
understanding of the interactions between different scales of turbulence. But unfortunately
we are very far from such an understanding. We therefore have to make interpretations and
guesses when we relate model results to the real world.
The Fuga model is based on the following interpretation of the situation. We will assume
that eddy viscosity is able to describe the effects of the smallest and the ’scrambling, eddies,
but not the larger ones. The main reason is that a RANS model does not ’know’ about large
eddies, unless the explicit mechanisms that create them are included in the model. Thus
if gravity waves create meandering eddies, the model should be able to deal with gravity
waves before we can assume that these eddies are incorporated in the model results. A
time resolved model would be needed for that, and RANS models are not (should not be)
time resolved. It would take an extremely complex model to reproduce weather on large
scale and the local effects if it on small scales, but in principle this can be done by nesting
models of progressively finer resolution. The demands for computing power for such a model
would be exorbitant. The finest level would probably have to be a LES model, and the next
level should be sophisticated enough to provide realistic boundary conditions for the LES
model that can reflect the effect of mesoscale eddies. A LES model alone is not enough
because it only transforms boundary conditions to bulk behaviour. In other words, even if
the LES model exhibit a great deal of physical realism, the results are no better than the
eddies we feed to it. We are looking for a much simpler solution. In our case we use the
simplest possible RANS model which certainly does not include any mesoscale behaviour,
and consequently we interpret the model results as representing the behaviour of wakes in
an idealized world where mesoscale eddies do not exist. They do exist in the real world,
but, to the extend it makes sense, our interpretation allows us to add mesoscale effects as a
kind of post processing, and we can add them without double counting. We should of course
make sure that the CFD model was not tuned to reproduce data that include mesoscale
effects, but that is not the case for the simple closure we use because it does not have any
adjustable parameters.
4.3 Wake meandering modelling

So far we have argued that meandering is an aspect of real wake behaviour that present day
models cannot cope with in a fully realistic manner. The old model is no exception. In the
new model meandering is to a large degree ’put in by hand’ as a simple post–processing of
model results. In section 4.1 we discussed the effect of slow variations of the wind direction
and showed how this and the effect of distance to the met mast could be treated by applying
a appropriate filter. This takes care of the variations shown as the red line in figure 1. We still
need to consider the ’meandering’ scales in figure 4.2 which are represented by the difference
between the green and the red lines in figure 1. The following analysis is based on wind data
from sonic anemometers sampled at 20Hz. The data was divided onto ten minutes long time
series each of which were de–trended by subtracting a best–fit straight line variation. Ideally
the red lines in figure 1 should have been used for the detrending, but we have checked on
a few examples that the two methods give the same results for all practical purposes.
The original plan was to install the machinery of the DWM model ( Larsen, Madsen,
Thomsen and Larsen 2008) into Fuga, but this proved more problematic than we had an-
ticipated. In the DWM model wake trajectories are generated by following passive fluid
particles in a simulated, frozen turbulent velocity field which is convected downwind with
an appropriate, uniform velocity. In order to simplify things further, the downwind com-
ponent is replaced by the constant advection speed, so that the fluid particle always stays
within the same cross–wind 2D slab of the frozen velocity field. The main problem is that
the wake trajectories generated in this way stop meandering after a while. They look nice
and wiggly–waggly at the start, but eventually they all become straight lines. This does
not happen right away, but only after a distance somewhat longer than the typical distance
between the turbines. The DWM model is intended for estimation of wake effects from tur-

bines located immediately upwind at a relatively short distance, where the trajectories are
still meandering. Used in this way the model results are not seriously affected, but when
it is used for larger distances, comparable to the overall size of a wind farm, the spurious
behaviour becomes a problem. The intention here was to model wake effects across wind
farms and even between wind farms, and therefore something had to be done.
The basic idea in the DWM model, and in Fuga, is that the momentum deficit in a wake
can be treated as if it were a passive tracer. In other words, the wake behaviour is similar
to a passive plume. In the near part of a wake, where the velocity in the wake centre is
much reduced compared to the surroundings, this is a crude assumption. However, further
downwind, where the wake is just a small perturbation, the assumption can be expected to
hold. The justification for treating the wake as a passive tracer is therefore similar to the
justification for the use of a linearised model, which we know works surprisingly well.
We have found that the cause of the odd behaviour is that the 2D flow field in the slab
containing the fluid particle is compressible. The simulated 3D flow is compressible so that
∂v ∂w ∂u
+ =− (36)
∂y ∂z ∂x
For the 2D flow, where u is taken to be constant, the right side will act as a source term. In
areas where it acts as a sink, the points where v = w = 0 will be attractors. The fluid particle
in the 2D flow therefore stops moving when it hits an attractor. After having realized this, it
was decided to use the full 3D incompressible flow to generate trajectories. This eliminates
the problem.
Meandering is implemented in the model using the kinematic simulation technique by
Mann (1998). The kinematic simulation is based on rapid distortion theory (cfr. Mann
1994) which takes the effect of shear into account. The method has lately been modified
by Peña, Gryning and Mann (2010) to incorporate modifications due to stability, and the
modified model was validated against data from the Hovsore test station. The result is
a machinery that can generate flow field examples which possess many of the statistical
properties observed in real, atmospheric turbulence.
0.6
0.4
fvHu*tL
0.2
0.0
0 20 40 60 80 100 120 140

u*t
Figure 7. Dots: normalized autocorrelation function of v obtained from simulated wind fields.
Red line: fit.
This simulation technique works well, but it uses much computer resources both in terms
memory and cpu time. It is therefore usually run on a cluster. but Fuga is supposed to run

on a single PC, so something had to be done in order to speed things up. The cure was
to approximate the trajectories with a Gaussian stochastic process, which is much easier
to simulate - making a model of a model so to speak. The simulated trajectories are non–
Gaussian, even if the simulated fields are Gaussian, but a Gaussian approximation can
reproduce the Lagrangian velocity autocorrelation function and therefore it will also get the
second moment of the crosswind deviation of the trajectory right ( Taylor 1921). A large
number of simulations were made for different stabilities (different values of 1/L) and velocity
auto correlation functions were made separately for each of the two cross wind components.
The simulated velocities were scaled with σu . If time is replaced with the variable σu t then
σu disappears from the problem and the autocorrelation functions can be expressed as
hv(t) v(t + τ )i = σu2 fv (σu τ ) (37)

hw(t) w(t + τ )i = σu2 fw (σu τ ) (38)
where fw and fw depend on L but not on the wind speed. The functions behave similarly
and a typical example is shown in figure 7. The uncertainty of fw is small for small time lags,
and increases as the time lag increases. The tail would therefore probably have approached
0 in a nicer way if more simulations were used, and the undershooting (negative values) in
this example is just noise. The red line is a fit of the form
fv (s) = µ2 (1 + λ|s|)e−λ|s| (39)
The expression fits all the data with values of µ and λ depending on stability. It is computa-
tionally inexpensive to generate a Gaussian process v̂(s) with this autocorrelation function
and obtain the deviation Z ∞
ŷ(s) = v̂(s0 ) ds0 (40)
0
A formal solution that gives (39) can be written as

Z ∞
v̂(s) = 2µλ3/2
s0 e−λs0 N (s − s0 ) ds0 (41)
0
where N is δ correlated Gaussian white noise, i.e

hN (s) N (s0 )i = δ(s − s0 ) (42)
Note that (43) is a solution to the equation
2
∂
λ+ v̂(s) = 2µλ3/2 N (s − s0 ) (43)
∂s
One can integrate this equation numerically or solve it using a Fourier transform (which
transforms white noise into white noise), but there is an even faster method that generates
ŷ(s) directly for a discrete set of s values. The simulation is therefore just a matter of putting
some random numbers into analytical expressions. The derivation is rather lengthy and is
therefore skipped.
Some simplifications were made: v̂ and ŵ are taken to be independent and we do not
include a û so that the fluid point moves downwind with constant velocity. This is convenient
because the time delay before the wake from one turbine hits the rotor of another turbine
is deterministic rather than stochastic.
We are using the same constants for the spectra as is used in the DWM model. They are
normalized with σu because the experimental spectra were normalized in this way. For our
purpose it would have been better if they were normalized with u∗ , but σu was preferred
because it is much more robust to measure than u∗ . According to Monin–Obukhov theory
σu /u∗ should be a function of ζ, but in reality it does not work. According to Panofsky and
Dutton (1984) it is better expressed as a function of zi /L for the unstable cases, σu /u∗ ∼ 2.4
for neutral and moderately stable cases while for strong stability the value becomes ’quite
large and unpredictable’. Hicks (1981) presents experimental data and draws a curve that

indicates that σu /u∗ saturates at a value of 3.0 for very unstable conditions. However, this
behaviour is not really supported by the scattered data. The value σu /u∗ = 2.4 for neutral
conditions is, on the other hand, supported by a large number of experiments (see Panofsky
and Dutton 1984), and because of the uncertainty for non–neutral conditions we have chosen
the value σu /u∗ ∼ 2.4 to account for all stabilities.
1.2
1.0
0.8
m @-D
0.6
0.4
0.2
0.0
-0.01 0.00 0.01 0.02 0.03
1•L @m D -1
Figure 8. The µ parameter as function of the inverse Monin–Obukhov length. Blue is for the
v component and green is for the w component
0.08
0.06
l @m-1 D
0.04
0.02
0.00
-0.01 0.00 0.01 0.02 0.03
1•L @m D-1
Figure 9. The λ parameter as function of the inverse Monin–Obukhov length. Blue is for the
v component and green is for the w component

5 Rotor averaging
Figure 10. The seven points used in the seven point method
The thrust is usually evaluated from the simple formula

T /ρ = 1/2 CT πR2 Uhub
2
(44)
This is valid in a constant wind profile without shear. In blade element theory the thrust
is calculated separately for blade elements using the local wind velocity. This indicates that
we might (but see below) get a slightly better approximation if we integrate over the rotor
disk: Z
T /ρ = 1/2 CT U 2 dA (45)
disk
2
The integral can be performed numerically to any desired accuracy, and Uhub πR2 can be
regarded as the first approximation:
Z R Z 2π
dr dθ F (r, θ) ≈ πR2 F (0, 0) (46)
0 0
A more accurate can be improved by using the following seven point method instead of a
one point method:
 
Z R Z 2π 5
1 1 X p
dr dθ F (r, θ) ≈ πR2  F (0, 0) + F ( 2/3R, jπ/3) (47)
0 0 4 8 j=0
The formula is valid if F is a polynomial in y = r cos θ and z = r sin θ with degree less
than 5.pIt is most practical to arrange the seven points in three different heights: zhub and
zhub ± 2/3R. The output from the model is restricted to certain levels that jump about
10% from level to level. An interpolation is made when evaluating U at the three desired
heights, which spoils some of the accuracy, but the method is still definitely better than the
one point method.
Thrust curves are normally calculated assuming a logarithmic profile or measured indi-
rectly by strain gauges placed on the tower. The thrust coefficient is then obtained using the

one point approximation. When the seven point method is used the resulting effective hub
height velocity tends to be larger than the actual hub height velocity, so that the estimated
power production increases. In order to stay on the measured (or calculated) thrust curve,
we therefore scale the results so that the seven point method gives the same results as the
one point method for the free flow profile. In this way the production from an unobstructed
turbine does not depend on which method is used.
The rotor average velocity used for the thrust is also used for the production. Ideally we
should come from the rotor average of U 3 , but the error introduced is negligible.
The seven point method is optional in the Fuga GUI. It should be noted, however, that
(45) we cannot simply use the CT values from an ’official’ thrust curve unless the blades
are individually pitch controlled. When the blades do not have individual control, the angle
of attack will change as a result of the shear as the blade completed a circle. When the
turbine operates below rated power, the angle of attack is optimized and any change from
the optimum value will decrease the performance. In that situation an increase of the wind
speed will in fact result in a decrease thrust and a decrease of production. Therefore it is
likely that the overall performance drops because of shear rather than increases. We therefore
emphasize that the seven point method should not be used for turbines without individual
pitch control.

6 Validation
Model evaluation addresses the question of justifying the ’correctness’ of a model. This could
be divided into three aspects:
1. Model tuning. Empirical basis for simplifications including justification of model con-
stants.
2. Model validation. Comparison of model predictions with measured data (different from
those used for 1).
3. Code verification. Checks that can justify that computer code is performing the tasks it
is supposed to do and that this is actually leading to a proper solution of the problem.
This report mainly describes how we will deal with model tuning and model performance.
In this project code verification has been informal consisting in various cross checks that can
be made along the way as the programs were developed. This can involve comparison with
known, analytical solutions to special or limiting cases, the making of a mirror program on
a different platform (e.g. Mathematica) and debugging in general. By now Fuga has into a
complex application as more and more features have been added, and the resources necessary
for debugging have been much larger than anticipated.
In this section we compare model results with experimental data. The original plan was
that Carbon Trust should provide test data, but no data was delivered. It is our impression
that various Carbon Trust partners have tested the model against data from a number
of offshore wind farms, but the results of these tests have not been shared with us. The
validation is therefore less complete than it could have been.
6.1 Repeats of old tests

The old model has been successfully validated against data from several offshore wind farms
and the new model was run for the same cases. As for the old model neutral stability
was assumed even if data was not necessarily sorted by stability. The new model generally
reproduces the results of the old model. Due to changes of numerical procedures the results
are not exactly the same. Figure 11 show a repetition of the validation against Horns Rev 1
production data using the new model. No filters were applied and the same input were used
as to the old model: U = 10m/s, ζ0 = 0 and z0 = 0.1mm. The same qualitative behaviour
is seen with an under–prediction of the efficiency for the narrowest bin. For the wider bins
there is an excellent agreement between the old model and data while the new model over–
predicts a few percent for the higher row numbers. In figure 11c both meandering, trend and
spatial filters are included. This clearly improves the agreement with data, especially for the
narrow ±2.5 degree bin, where the ’raw’ results vastly under–predict the data. This changes
when meandering and the filters are applied. Of these the spatial de–correlation filter has
the largest impact. The results change from under–prediction to a slight over–prediction.
This could be caused by the way data were screened. The data represent ten minute periods
where the averages of the preceding period fall into the same wind speed and direction bins.
The idea was to select as clean cases as possible, but even if the screening might have some
of that effect, the model results suggest that a large degree of uncertainty remain. A separate
analysis of the met data to adjust to this particular screening procedure has not been made,
but the result would have been a narrower filter that would bring model results and data
closer together.

1.00
270+-7.5 Fuga
270+-12.5 Fuga
0.90
270+-17.5 Fuga
270+-7.5 Data
Normalized prod. 0.80 270+-12.5 Data
270+-17.5 Data
0.70
0.60
0.50
0.40
1 2 3 4 5 6 7 8 9 10
a Row number
1 270+-2.5 Fuga
270+-7.5 Fuga
270+-12.5 Fuga
0.9
270+-17.5 Fuga
270+-2.5 Data
0.8 270+-7.5 Data
Normalized prod.
270+-12.5 Data
270+-17.5 Data
0.7
0.6
0.5
0.4
1 2 3 4 5 6 7 8 9 10
Row number
b
1 270+-2.5 Fuga
270+-7.5 Fuga
270+-12.5 Fuga
0.9
270+-17.5 Fuga
270+-2.5 Data
0.8 270+-7.5 Data
Normalized prod.
270+-12.5 Data
270+-17.5 Data
0.7
0.6
0.5
0.4
1 2 3 4 5 6 7 8 9 10
Row number
c
Figure 11. Comparison of Fuga results with data from the Horns Rev I wind farm. U=10m/s,
z0 =0.1mm, a: the old model. b: new model without meandering or filter. c: new model with
meandering and filter.

6.2 New Horns Rev 1 test
Table 2. Stability classes used for the test and input used for the 9±1m/s bin.
Description Class L interval h1/Li−1

data z0 ζ0
−4 −6
Very unstable -3 -100m< L <-50m -72m 1.15 10 m -1.58 10
Unstable -2 -200m< L <-100m -137m 1.09 10−4 m -8.00 10−7
Near–neutral unstable -1 -500m< L <-200m -285m 102 10−4 m -3.62 10−7
Neutral 0 |L| >-500m -9257m 9.37 10−5 m -1.44 10−8
Near–neutral stable 1 200m< L <500m 291m 8.24 10−5 m +2.72 10−7
Stable 2 50m< L <200m 88m 5.77 10−5 m +6.50 10−7
Very stable 3 10m< L <50m - - -
A new test was prepared for the Horns Rev 1 data set. AMOK was used to extract stability
information from met mast M7 data together with SCADA data for turbine production. The
wind direction was measured by a wind vane (z = 60m) and the velocity was measured by
a cup anemometer (z = 70m) both on the met mast. M7 is located 6km to the east of the
farm, which is further away from the farm than both M2 and M6, but the data from M7
are more reliable. Data was binned according to wind speed (2m/s bins), wind direction (5
degree bins) and inverse Monin–Obukhov length binned in the classes defined in Gryning
et al. (2007), see Table 2. The table also contains the bin averages used for model input. Data
for U = 9 ± 1 m/s were selected for evaluation. The very stable cases were omitted. Only
wind directions where the met mast was clear of the wake from the farm were used. Wind
directions between N and NE are very rare, hence useable wind directions are limited to bins
between 45 degrees and 240 degrees. Production data were selected for each turbine and data
were rejected unless all upwind turbines were running. The average turbine production was
therefore based on different data for each turbine since down wind turbines were allowed
not be running. The production efficiency was calculated using the average production of
all the running, unobstructed turbines as reference. For some wind direction bins there were
little data or none at all. For an individual turbine the spread of the production data in a
bin is about 30% 3 . With between 10 add 100 realizations the error of the mean value is on
the order of 3–10%. The error on the production efficiency is dominated by the estimate of
the undisturbed production, which consists of an average over 8–18 turbines depending on
the wind direction. The uncertainty of this average is lower that for a single turbine, but it
is difficult to say how much lower because the productions from the individual turbines are
correlated. There are also systematic errors such as spatial and temporal changes of wind
speed and stability and coastal effects making the climate inhomogeneous. This suggests an
accuracy of the data of perhaps 5%.
The model was run for all the stability classes except class +3, where the criterion that
z/L < 1 is violated. Three different settings were used: 1) ’raw’ results without meandering
and filtering, 2) with meandering turned on and 3) with both meandering and trend and
spatial filtering. The one point method was used in all cases.
Figure 12 shows results for Class 0 (neutral). The effect of the stochastic meandering can
be seen, but it is not large. This wold perhaps change if the measured σu was used as input
(the model uses the relation σu /u∗ = 2.4). The trend and spatial de–correlation filter has
more impact, perhaps a little bit too much in this case. The overall agreement with data is
fair even if the model over–predicts the efficiency by about 3%.
Figure 13 shows results with both meandering and filtering for all stability classes (except
very stable). The general picture is again that the model over–predicts by 2–5%.
Table 3 shows the mean value of the relative production for those wind direction bins
where enough data exists to allow an estimate of the total production to be made. The
3 All percentages in this section are relative to the undisturbed reference production

model results and data in table 3 are therefore within the error bars and the discrepancy
could therefore possibly be due to data uncertainty. However, the consequent over–prediction
could also very well be for real. The numbers cannot be compared between stability classes
because the amount of available data differs slightly. In order to compare we have therefore
also computed the predicted efficiency averaged over all directions for all seven stability
classes. For class +3 (very stable) we have used L = 37m corresponding to ζ0 = 1.0 10−6 .
The value zhub /L ∼ 2 is beyond the range where the M–O theory is valid, but we have tried
to run the model anyway. Figure 14 shows the somewhat discouraging results. Something
is happening in the stable (class +2) and very stable (class +3) the makes the efficiency
increase with stability. We cannot prove it, but we believe that this odd behaviour is caused
by the numerical problems that were mentioned in section 3.3.
It should be noted, that these tests are not controlled experiments because we are testing
two things at the same time. On the one hand we have the model of the uncertainty or
imprecision of the measured mean values, and on the other hand we have the wake model.
Discrepancies between model and data can likewise have more than one explanation. The
width and depth of the dip of the efficiency around a direction where turbine columns are
aligned with the wind, depend both on the predicted wake width and on the uncertainty we
assigned to the wind direction measurement. If the dip is too narrow, say, then it could be
because the model predicts a wake that is too narrow, but it could also be because we have
underestimated the uncertainty of the wind direction. A more controlled experiment, where
a discrepancy only has one explanation, should involve direct measurements of the wake.
Table 3. Mean farm efficiency over the sector 42.5–242.5 degrees where data exist for all
turbines.
Class -3 -2 -1 0 +1 +2
Modelled efficiency 87.2% 86.3% 85.3% 83.1% 81.3% 80.7%
Observed efficiency 83.7% 82.4% 80.8% 79.9% 79.0% 76.0%
difference 3.5% 3.9% 4.5% 3.2% 2.3% 4.7%

a
0 30 60 90 120 150 180 210 240 270 300 330
1. 1.
0.8 0.8
Farm Efficiency
0.6 0.6
0.4 0.4
0.2 0.2
0. 0.
0 30 60 90 120 150 180 210 240 270 300 330
q
b
0 30 60 90 120 150 180 210 240 270 300 330
1. 1.
0.8 0.8
Farm Efficiency
0.6 0.6
0.4 0.4
0.2 0.2
0. 0.
0 30 60 90 120 150 180 210 240 270 300 330
q
c
0 30 60 90 120 150 180 210 240 270 300 330
1. 1.
0.8 0.8
Farm Efficiency
0.6 0.6
0.4 0.4
0.2 0.2
0. 0.
0 30 60 90 120 150 180 210 240 270 300 330
q
d
0 30 60 90 120 150 180 210 240 270 300 330
1. 1.
0.8 0.8
Farm Efficiency
0.6 0.6
0.4 0.4
0.2 0.2
0. 0.
0 30 60 90 120 150 180 210 240 270 300 330
q
Figure 12. Red dots: observed total farm efficiency for Horns Rev 1. Neutral (C=0) and U =
9 ± 1m/s. Blue line: model predictions. a: without filters. b: meandering filter. c: meandering
and trend filter. d: meandering and trend and spatial filter

0 30 60 90 120 150 180 210 240 270 300 330
1. 1.
0.8 0.8
Farm Efficiency
0.6 0.6
0.4 0.4
0.2 0.2
0. 0.
0 30 60 90 120 150 180 210 240 270 300 330
0 30 60 90 120 150
q
180 210 240 270 300 330
Class -3
1. 1.
0.8 0.8
Farm Efficiency
0.6 0.6
0.4 0.4
0.2 0.2
0. 0.
0 30 60 90 120 150 180 210 240 270 300 330
0 30 60 90 120 150
q
180 210 240 270 300 330
Class -2
1. 1.
0.8 0.8
Farm Efficiency
0.6 0.6
0.4 0.4
0.2 0.2
0. 0.
0 30 60 90 120 150 180 210 240 270 300 330
0 30 60 90 120 150
q
180 210 240 270 300 330
Class -1
1. 1.
0.8 0.8
Farm Efficiency
0.6 0.6
0.4 0.4
0.2 0.2
0. 0.
0 30 60 90 120 150 180 210 240 270 300 330
0 30 60 90 120 150
q
180 210 240 270 300 330
Class 0
1. 1.
0.8 0.8
Farm Efficiency
0.6 0.6
0.4 0.4
0.2 0.2
0. 0.
0 30 60 90 120 150 180 210 240 270 300 330
q Class +1
0 30 60 90 120 150 180 210 240 270 300 330
1. 1.
0.8 0.8
Farm Efficiency
0.6 0.6
0.4 0.4
0.2 0.2
0. 0.
0 30 60 90 120 150 180 210 240 270 300 330
q Class +2
Figure 13. Red dots: observed total farm efficiency for Horns Rev 1 with ±2.5 wind direction
bins. Blue line: model predictions for U = 9 ± 1m/s with 4 meandering repeats per direction,
±2.5 degree rectangular filter and filter for trend and spatial de–correlation. Stability class
ranges from -3 (upper) to +2 (lower)

0.90
0.88
0.86
360o mean efficiency
0.84
0.82
0.80
0.78
0.76
-1.5 ´ 10-6 -1. ´ 10-6 -5. ´ 10-7 0 5. ´ 10-7 1. ´ 10-6

z0
Figure 14. Relative production averaged over 360 degree for the seven stability classes.

7 Guidelines
7.1 Stability input to Fuga
We mentioned earlier that ζ ≡ z/L is a local measure of stability, while the local inverse
Monin–Obukhov length 1/L can be regarded as a global measure for the whole (surface)
layer. In Fuga the basic inputs are: wind speed U at hub height, wind direction, boundary
layer depth zi , roughness z0 and the new parameter ζ0 ≡ z0 /L. We recommend always to
use zi ≈ 400m, which leaves us with two parameters: z0 and the ζ0 . On land the roughness
is usually regarded as a constant independent of L and often also independent of U , but the
roughness of a sea surface depends on wind speed. More precisely, z0 depends on u∗ because
the momentum flux at the surface is responsible for the excitation of the waves that make
the surface rough. Using (2) and (13) (with Bc = 0 for simplicity) we find
ζ0 = Ac κ θ∗ /T (48)
In other words, ζ0 is essentially a measure of θ∗ , the temperature scale of the layer. This
is an indicator of stability, but a different one from e.g. 1/L, which is sometimes used. For
model evaluations and Annual Energy Production (AEP) estimates it is important to enter
the right ζ0 . The AMOK tool can be used to convert met mast measurements to estimates
of z0 , L and ζ0 .
Monin–Obukhov theory is not reliable for extremely stable conditions where profiles es-
sentially become unpredictable, and the diabatic functions φm and φh are ill-determined for
values of ζ larger than 1. We recommend not to use trust results for cases where L is smaller
than the hub height, because the underlying theory is not valid.
The shadowing is generally more severe in stable conditions and less severe in unstable
conditions compared to neutral, and things tend to even out when the AEP is estimated,
However, even a small difference could be decisive for the choice of the location of a wind
farm. Unfortunately, the numerical solutions break down for too stable cases and yields
results that we are far too optimistic with respect to wake losses. The present solver4 is not
reliable for ζ0 & +3 10−7 . This corresponds to L ∼ 130m or zhub /L ∼ 0.5.
7.2 Roter averaging

Rotor averaging using a one point and a seven point method was explained in section 5. We
recommend to use the one point method and only use the seven point method if the blades
are individually pitch controlled.
7.3 How to make a climatology with stability

Lack of local met data is often a problem, and it only becomes worse when an extra di-
mension is added to the wind climate. When stability is not taken into account, the data
are usually binned along two dimensions: wind speed and wind direction. With about 100
bins and 100000 data points (corresponding to about two years of ten minutes averaged
data)
√ the typical bin will contain about 1000 counts which gives an uncertainty of about
1/ 1000 . 3%. Adding a stability dimension we may √ end up with 1000 bins containing
about 100 counts each now with an uncertainty of 1/ 100 . 10%, which may or may not be
acceptable. For model evaluations the situation gets even worse because many data points
have to be rejected due to turbines not running, bad data quality etc. One could also be
tempted to use small wind direction bins in order to try to catch the sensitivity to this
parameter. 5 degree bins are often used corresponding to 72 bins along the wind direction
dimension alone. As a result we may end up with a lot of half empty bins unless several years
4 Preludium-T 16/09/2013 12:47

of meteorology and production data are available. Reanalysis data from mesoscale models
could be a means of filling the data gap. This has proved useful for wind speed and direction
climatology (see Hahmann, Lange, Pena and Hasager 2012) and it might be possible to add
the stability dimension this way.
The data sets from the met masts at Horns Rev are large enough to make a climatology
analysis. We will use data from mast M7 located to the East of Horns Rev 1 to illustrate
how a climatology can be made.
First the data is divided into three groups: calm, stable and unstable data. The calm
data are all data with wind speeds less than 3m/s. Even if the meteorology at low wind
speeds could be interesting, 3m/s is below the cut–in speed of the turbines, so no turbines
are running. It is natural to split the rest into stable and unstable cases even it means that
the neutral part is split in two. Stable and unstable cases are treated separately and the
exercise consists in estimating a (conditional) probability density function (pdf) which is a
function of three variables: The wind direction α and two parameters somehow representing
wind speed and stability. The two parameters should be chosen so that they determine the
velocity and temperature profiles. In other words, we should be able to work out the values
of u∗ , θ∗ , z0 and L from them, but otherwise we can do as we please. The traditional choice
is U at some height and 1/L. However, U and 1/L are strongly correlated which makes it
difficult to divide the (U ,1/L) plane into bins each containing about the same amount of
data. We will choose u∗ and ζ0 , because they are close to being statistically independent. It
is also straight forward to use (13) to generate the Fuga input z0 .
The general problem is to estimate mean values like
Z Z Z
hF (u∗ , ζ0 , α)i = du∗ dζ0 dα P (u∗ , ζ0 , α)F (u∗ , ζ0 , α) (49)
where P (u∗ , ζ0 , α) is the joint pdf of u∗ , ζ0 and α (the wind direction), and F (u∗ , ζ0 , α)
is a function such as the energy production. The conventional method is to divide the
approximate (u∗ , ζ0 , α) space into bins and approximate F by a function which is constant
over each bin so that
Z Z Z X X
hF (u∗ , ζ0 , α)i ≈ du∗ dζ0 dα P (u∗ , ζ0 , α) Fbin Ξbin (u∗ , ζ0 , α) = Fbin hξbin (u∗ , ζ0 , α)i
bins bins
(50)
where {ξbin } is the set of characteristic functions for the bins. The method works fine as
long as we have enough bins to give a decent representation of F . In our case we need to
make quite large bins in order to avoid that they become half empty, and the approximation
could become inaccurate. We will therefore try to improve the approximation by the use of
chapeau functions. It is best to illustrate the idea with a 1D distribution instead of a 3D.
A chapeau function ∆ is a piecewise linear function of the form

 0 for x < a
 x−a for a < x < b

∆(x) = b−a (51)
b−x

 c−b for b < x < c

0 for c < x
We divide the x–axis into segments by a set of points {xi } and define chapeau functions ∆i
using (51) with (a, b, c) = (xi−1 , xi , xi+1 ). We can use the chapeau functions to approximate
a function F (x) by a continuous, piecewise linear function, viz.
X
F (x) ≈ F (xi ) ∆i (x) (52)
i
Since ∆i (xj ) = δij , the right hand side attains the right values for x = xi and is linear in
between. This is a slightly better than the piecewise constant approximation. For the mean
value we get

Z X X
hF (x)i ≈ dx P (x) F (xi ) ∆i (x) = F (xi ) h∆i (x)i (53)
i i
The 3D case is similar. We use chapeau functions for the variables u∗ and ζ0 , divide the
two axes at points {u∗i } and {ζ0j } and define 2D chapeau functions as the product of two
1D chapeau functions:
∆ij (u∗ , ζ0 ) ≡ ∆i (u∗ ) ∆j (ζ0 ) (54)
It is natural to use Fourier modes for the wind direction and write F as
∞
X
F (u∗ , ζ0 , α) = Ck (u∗ , ζ0 ) eikα (55)
k=−∞
with α expressed in radians. We then use the approximation

∞
XX X
F (u∗ , ζ0 , α) ≈ ∆ij (u∗ , ζ0 )Ck (u∗ , ζ0 ) eikα φk (56)
i j k=−∞
where φk is a filter that cuts off the infinity sum over k and helps blurring the data. Using
this approximation for F we get
X X
hF (u∗ , ζ0 , α)i ≈ Ck (u∗i , ζ0j ) h∆ij (u∗ , ζ0 ) eikα i φk (57)
ij k
The analysis outlined above was carried out for data from the years 2005-2009. The dataset
contains about 240000 records of ten minutes averages. Wind speed at 20m, temperature at
16m, water temperature 1m below sea level, and wind direction at 68m were used to find
1/L, z0 , θ∗ , and u∗ . Figure 15 is a sanity check where the wind speed at 70m was estimated
est est
using (5) and compared to measurements. The values of the ratio U70 /U70 scatter nicely
around 1.0 with only a small bias of about 2 percent. It is typical that the scatter if largest
in the stable region with ζ = 70m/L > 0.5.
Figure 15. The windspeed extrapolated from 20m to 70m compared to the measured wind
speed at 70m. The green line is the average.

7.4 Fuga user interface
The software consists of updated versions of the programs described in Ott et al. (2011).
The linearized CFD model, implemented in the Preludium-T program, stores solutions in
compact look-up tables. The Trafalgar program extracts normalized velocity deficits behind
a single turbine. The Fuga program aggregates single-wake solutions to wind-farm wakes and
displays result tables and plots in a graphical user interface. Fugabatch is similar to Fuga,
except that it is called from command line and scripts and has no graphical user interface.
The file structure containing look-up tables and Trafalgar output is nearly unchanged,
except that we now have an extra dimension of atmospheric stability. As before, we have
lookup-tables specific for each turbine type (*.lut files) and more general pre-lookup tables
(*.pre files). File names are extended with a reference to the stability parameter ζ0 . Previ-
ously, a set of general look-up tables for neutral atmospheric stability was calculated once
and for all in the first Fuga session. This basic calculation is now repeated each time a case
with a new stability parameter is specified.
Figure 16. Turbine site selection in Fuga with a) all sites included, b) only sites in the last
wind farm included, and c) all sites included except the first wind farm and phase 2 of the
last wind farm.
The wind-farm layout is specified by a WAsP project file, which includes turbine positions
and heights, site-specific wind climates, and power curves with associated thrust-coefficient

curves. Turbine sites can be organised in nested site groups with different turbine types
associated. Figure 16 shows a wind-farm cluster combining three wind farms with individual
turbine types. The display in Fuga is similar to that of WAsP, except that three objects
names are emphasized with bold characters. These selected objects indicate 1) a group of
turbines chosen for analysis, 2) a reference site for specifying wind conditions, and 3) a
turbine type selected for single-wake displays. You modify this selection by clicking on the
object tree. Figure 16b shows a situation where the turbine site selection in is reduced to
only one wind farm. You can also temporarily deactivate turbine sites and site groups by
right-clicking them the object hierarchy, as shown in Figure 16c. These editing methods
enable the user to calculate the production of a wind farm power with and without the
effect of wakes from neighbouring wind farms.
The reference site is a position where you specify the wind speed and direction. This could
be a met mast and anemometer height or a turbine site and hub height. The free wind speed
at a turbine site will differ from that of the reference site if the height differ. Furthermore, the
distance between a turbines and reference site will affect the spatial averaging filter on the
directional variation of the power production, since the correlation between wind directions
at two sites decreases with spatial separation.
Figure 17. Screen dump of the Fuga user interface.
Figure 17 shows the user interface with the atmospheric boundary layer defined in the
lower-left area. The wind speed and direction specified here applies to the selected reference
site. The case list refers to existing linearized CFD solutions, and you may either select one
of these or specify a new combination of z0 , zi and ζ0 . You can use the separate AMOK
program to derive these parameters from wind speed, air temperature and sea temperature.
Methods for directional averaging are chosen in the top-right area of the user-interface
window. Some of these have parameters, like the number of repetitions for wake meander
simulations or the spread of a Gaussian averaging filter. Options for modelling effects of shift-
ing 10-min averaged wind directions and spatial de-correlation between the wind direction
at reference site and turbine sites are always available for the wake meander method. The
simple box averaging and Gaussian averaging filters are available when calculating annual
energy production.

The area under the averaging options has differnt pages. The first page called ’file overview’
gives a survey of existing look-up tables and wake velocity fields. It also includes a frame
showing output of the preludium and Trafalgar programs, from where you can follow cal-
culation progress and inspect log files from previous calculations. It is a good idea to check
these log files in case Fuga fail to load wake data.
The second page called ’turbine site list’ provides turbine powers for a given wind situation
and directional averaging method. Aggregated powers for wind turbine groups are shown
at the end of this list. These results include wake effect of neighbouring wind farms unless
these are deactivated from the object hierarchy as shown in Figure 16.
The third page called ’production’ shows a similar list of annual energy production calcu-
lated with the turbine-site specific wind climates imported with the WAsP project. As an
approximation you may calculate this for a boundary-layer case with a typical combination
of surface roughness, inversion height and stability. A better way is to use a probability-
weighted mixture of cases with statistical variation of atmospheric stability and with sur-
face roughness depending on wind speed and stability. You may specify such a variation for
Fuga, but we have not yet developed a method for linking it to a climatology of the form
hF (u∗ , ζ0 , α)i or optimizing the number of cases needed for reliable production estimates.
Figure 18. Wake behind a single turbine
The fourth page, shown in Figure 18, illustrates the wake behind a single turbine. The
coloured plot at the top shows horizontal velocity fields and the chart below shows wind
speed along crosswind, downwind or vertical transect lines defined by the user. Use this
plot type to check linearized CFD solutions or compare solutions for different atmospheric
stability.
The fifth page in Figure 19 displays the combined wake of all turbines with a wind-
speed profiles along a user-defined transect line. In the shown situation, we have chosen a
wind direction aligned with a row of turbines and defined a transect line passing through
the turbine positions. For the wake aggregation the program sorts the turbines after the
projected distance along the wind direction and starts by finding the thrust of the most
upwind turbine. This thrust and the single-wake solution is then used to find the reduced

Figure 19. The combined wake of multiple turbines.
wind speed at the next turbine position. The algorithm progresses in the downwind direction
until the thrust is known at all turbines, and then it is possible to lookup the reduced velocity
at any position. Lateral turbine displacements, wake centre-line displacement due to wake
meandering, rotor averaging, variable hub heights, and variable thrust-coefficient curves are
taken into account in this calculation.
Figure 20. Wake velocity profiles under unstable (L=100m), neutral and unstable (L=-100m)
atmospheric stability.
The lower panel can also display 2D profiles of wind speed or velocity deficit. The plots
in Figure 20 are combined of three such plots for variable atmospheric stability with a wind
direction only transect line only passing a single turbine. The eddy diffusivity increases with
height and especially in the unstable case we see that the wake is mainly depleted from the
top, so the height of the maximum velocity deficit decreases with distance. The downwind

wake development is very slow in the stable case.
Figure 21. Normalized power production of the Horns Rev I wind farm at 9 m/s as func-
tion of direction with a) no directional averaging, b) wake meandering, trend in wind speed
and spatial de-correlation of wind direction relative to the centre of the farm, and c) wake
meandering, trend and de-correlation relative to the M7 mast 6 km east of the farm.
Figure 21 shows a polar plot of aggregated power of the Horns Rev I wind farm at a
wind speed of 9 m/s normalized with a hypothetical production without wake effects. The
directions with minimal production efficiency are the ones where the wind blows along rows
of turbines. The three plots show the importance of directional filtering and the distance to
the reference site.
All tables and plots shown in Fuga may be saved to file or copied to the Windows clipboard
via the Fuga file menu. Tables are copied in tab-delimited text format, which is easy to paste
into Excel notebooks. It is also possible to invoke the result table menu and create flexible
tables of wake-reduced wind speeds or power as a function of wind speed and direction with
optional rotor averaging, directional averaging and wake meander simulation. Results are
calculated for individual turbine sites and for wind turbine groups.
The FugaBatch program is a non-visual tool, which do the same calculations as Fuga and
exports these to file. It is useful for sensitivity studies with systematic variation of model
input. The scripting syntax is explained in the Fuga help file and a collection of sample
scripts demonstrate typical uses.

References
Ayotte, K. W., Davy, R. J. and Coppin, P. A.: 2001, A simple temporal and spatial analysis
of flow in complex terrain in the context of wind energy, Boundary–Layer Meteorology
98, 275–295.
Barthelmie, R., Hansen, K. S., Frandsen, S. T., Rathmann, O., Schepers, J. G., Schlecz, W.,
Philips, J., Rados, K., Zervos, A., Politis, E. S. and Chaviaropoulos, P.: 2009, Modelling
and measuring flow and wind turbine wakes in large wind farms offshore, Winf Energygy
12, 431–444.
Charnock, H.: 1955, Wind stress on a water surface, QJRMS 81, 639–640.
Courtney, M. S. and Troen, I.: 1990, Wind speed spectrum from one year of continuous 8
hz measurements, Ninth Symposium on Turbulence and Diffusion, April 30 – May 3,
Risoe, Roskilde, Denmark, American Meteorological Society, pp. 301–304.
Grachev, A. A. and Fairall, C. W.: 1996, Dependence of the monin-obukhov stability pa-
rameter on the bulk richardson number over the ocean, Journal of Applied Meteorology
36, 406–414.
Gribben, B., de Villiers, P., Hui, S.-Y., Balcombe, C., Harrington, G., Peace, C., Mallins, D.,
Housley, P., Kristensen and Bech, K.: 2011, Wake effects developments in the offshore
wind accelerator, EWEA Wind Resource Assessment Technology Workshop,10–11 May,
Brussels, Belgium, EWEA.
Gryning, S.-E., Batchvarova, E., Brummer, B., Jørgensen, H. and Larsen, S.: 2007, On
the extension of the wind profile over homogeneous terrain beyond the surface layer,
Boundary-Layer Meteorology 124, 251–683.
Hahmann, A. N., Lange, J., Pena, A. and Hasager, C. B.: 2012, The norsewind numerical
wind atlas for the south baltic, Technical Report Report I-0011, DTU Wind Energy.
Hicks, B. B.: 1981, An examination of turbulence statistics in the surface boundary layer,
Boundary-Layer Meteorol. 21, 389–402.
Högström, U.: 1988, Non–dimensional wind and temperature profiles in the atmospheric
surface layer: A re–evaluation, Boundary-Layer Meteorology 42, 55–78.
Lange, B., Larsen, S., Højstrup, J. and Barthelmie, R.: 2004, The influence of thermal
effects on thewind speed profile of the coastalmarine boundary layer, Boundary-Layer
Meteorology 112, 587–617.
Larsen, G. C., Madsen, H. A., Thomsen, K. and Larsen, T. J.: 2008, Wake meandering: A
pragmatic approach., Wind Energy 11, 377–395.
Larsén, X. G., Vincent, C. and Larsen, S.: 2012, Spectral structure of mesoscale winds over
the water, Q. J. R. Meteorol. Soc. 139, 685–700.
Lindborg, E.: 1999, Can the atmospheric kinetic energy spectrum be explained by two-
dimensional turbulence?, J. Fluid Mech. 388.
Mann, J.: 1994, The spatial structure of neutral atmospheric surface-layer turbulence, J.
Fluid Mech. 273, 141–168.
Mann, J.: 1998, Wind field simulation, Prob. Engng. Mech. 13(4), 269–282.
Monin, A. S. and Obukhov, A. M.: 1954, Basic laws of turbulent mixing in the atmosphere
near the ground, Tr. Akad. Nauk. SSR 151, 163–187.
Monin, A. S. and Yaglom, A. M.: 1975, Statistical Fluid Mechanics, Vol. 2, The MIT Press.

Ott, S.: 2011, Linearized cfd, Technical Report Risø–I–3093(EN), Risø-DTU.
Ott, S., Berg, J. and Nielsen, M.: 2011, Linearized cfd models for wakes, Technical Report
Risø–R–1772(EN), Risø-DTU.
Panofsky, H. A. and Dutton, J. A.: 1984, Atmospheric Turbulence, John Wiley & Sons, New
York.
Pasquill, F.: 1974, Atmospheric diffusion, Ellis Horwood Limited.
Peña, A. and Gryning, S.-E.: 2008, Charnock’s roughness length model and non-dimensional
wind profiles over the sea, Boundary-Layer Meteorology 128, 191–203.
Peña, A., Gryning, S.-E. and Mann, J.: 2010, On the length-scale of the wind profile, Quar-
terly Journal of the Royal Meteorological Society 136, 2119–2131.
Sathe, A., Gryning, S.-E. and Peña, A.: 2011, Comparison of the atmospheric stability and
wind profile climatology at two wind farm sites over the north sea, Wind Energy . doi:
10.1002/we.456.
Taylor, G. I.: 1921, Diffusion by continuous movements, Proc. Roy. Soc. A224, 487.
Trust, C.: 2011. Scope of Work according to the contract.
Vincent, C. L., Larsén, X. G., Larsen, S. E. and Srensen, P.: 2013, Cross-spectra over the
sea from observations and mesoscale modelling, Boundary–Layer Meteorology 146, 297–
318. DOI: 10.1007/s10546-012-9754-1.

8 Appendix A: The Fuga help system
The following pages contain a printout of the Fuga help system.

Welcome to Fuga 2
The purpose of Fuga.exe is to demonstrate the method of linearized wake computations by lookup tables (LUTs). Fuga
calls the programs preludium.exe to generate preLUTs and LUTs and trafalgar.exe to calculate single-wake pro-
files. In addition Fuga calculates the combined wake from all turbines in a wind farm and estimates the annual power
production by a method similar to WAsP, however, with wake losses calculated by the Fuga wake model.
l Setting up calculations in Fuga 2
l Result views
l Further information
New in Fuga 2.5
Bug fix:
l Parameters in the wake-meander simulation model were not applied with the correct stability dependence.
l The filtering of the wind-direction dependence of the power output is now calculated as the last step. This cor-
rects a previous method where power was calculated from a filtered wind speed. The correction affects the
'wind direction trend' and 'spatial de-correlation' filters, but not the 'simple average' or 'Gaussian average'
filters.
New features:
l Improved control over wind direction filters modelling sampling effects of power statistics. The filter is pro-
vided in two versions - one for trends in the wind direction signal during sample periods and one for trend
plus statistical de-correlation due to spatial separation between reference mast and turbine sites. The filters
are not only available for the meander-model results but also for the top-hat and Gaussian averaging filters.
l FFT-based filters are also available from the 'Files| save results| result table' dialogue.
l Import of WAsP 11 workspaces.
New in Fuga 2.4
Bug fix:
l AEP calculations sometimes went wrong for projects using turbines with multiple power curves inside the
WAsP workspace, e.g. intended for different air densities. By mistake earlier versions of Fuga always used the
first power curve regardless of the WAsP workspace choice.
l Yet another correction for projects with turbines of different heights. A previous error, corrected in Fuga 2.2,
was incorrect reference velocity at the upwind turbine . The additional error, corrected in Fuga 2.4, was that
the velocity deficit at the downwind turbine sometimes was evaluated at the height of the wrong turbine.
New features:
l Wake meander effects
l Thrust and power optionally evaluated by a rotor-averaged inflow velocity
l Rotor averaging, variable wind direction, and wake meander now also affects AEP calculations and efficiency
for direction plot. Earlier versions only used directional averaging when calculating power performance for a
selected wind case. See how to set averaging options in Fuga GUI and in FugaBatch
l Alternative win64 versions of Fuga and Fugabatch allows Windows 7 users to use more RAM memory.
New in Fuga 2.3
New features:
l Single-turbine and windfarm wake effects also for boundary layers of non-neutral atmospheric stability
l Annual Energy Productions for a specified distribution of atmospheric stabilities
l Graphical display of specified stability statistics
See how to specify non-neutral stability in Fuga GUI and in FugaBatch
New in Fuga 2.2
Bug fix:
l Corrected AEP calculation for projects with turbines of different height. This problem existed in versions 2.0-
2.1 of both main Fuga and FugaBacth - but only with AEP calculations, not calculations for specific wind
cases.
l Corrected selection of tables in wind turbine generator files (*.wtg) with multiple performance tables or equiv-
alent multi-table turbine data in workspace files (*.wwh). The first table in the file was always selected erro-
neously ignoring user preferences.
New features:
l Result table may also be calculated by FugaBatch
l Adjustable colour schemes for single-wake and windfarm wake plots
l Graphical display of turbine-specific wake losses
l Output from preludium.exe and trafalgar.exe redirected to log files
New in Fuga 2.1
Bug fix:
l Corrected spreading in Gaussian filter used for directional averaging
New features:
l Speed optimizations in GUI - mainly by avoiding repeated calculations and interrupting time-consuming plot-
ting
l Optional speed and length scales plus coastline in the windfarm view
l Optional deactivation of sites or turbine groups by right-clicking objects in the workspace hierarchy
l Screening of incorrect far-field flow results due to lack of single-wake data
l Optional coarse-resolution plotting in wind-farm view (which is faster)
l Optional redirection of FugaBatch output to log file
New in Fuga 2.0
New features:
l Support for complex projects with multiple turbine types and variable turbine heights
l Import of WAsP workspace files, also in scripts
l Aggregated results for all turbine groups including nested subgroups
l Zoom in single-wake plot
New in Fuga 1.5
Bug fix:
l Data scanning routine might attempt to access PreLUT information before available. This was probably only a
problem when manually moving content of LUTs folders and thereby changing the Windows file order.
New features:
l Support for 3D wakes, however only in projects with a single turbine type
l Display of vertical wind profiles
New in Fuga 1.4
Bug fix:
l Resolutions of AEP calculations are now identical in Fuga and FugaBatch
New features:
l FugaBatch input is slightly simplified. There is no need to specify wake *.par and *.dat files as these are now
detected automatically.
l Possible to control the resolution for the AEP integral via parameters in fuga.ini
New in Fuga 1.3
Bug fix:
l Arrows next to wind direction and speed input boxes now results in a new calculation
New features:
l New versions of the preludium and trafalgar programs with minor modification to input files. Results are now
produced at the exact hub height.
l Now possible to average the calculated power and wind speed over a range of wind directions.
l A new data export facility is included.
l FugaBatch - a non-interactive mode for scripting of Fuga calculations
New in Fuga 1.2
Bug fixes:
l Improved response to change of wind direction
l Improved data export
l Improved wake modelling by much better versions of the preludium and trafalgar programs
New features:
l Precise positioning of transects in single-wake view
l Wind profile along transect in windfarm-wake view
l Zoom and pan in windfarm-wake view
New in Fuga 1.1
New features:
l First version with a help file
l Import from WAsP windfarm file (including local wind climates)
Setting up calculations in Fuga 2
Fuga 2 needs
l Wind farm layout and site-specific wind climate from a WAsP workspace file
l A boundary layer description.
Available data are presented in drop-down list on the left-hand side of the program window. In the first session these
lists will be empty and the user must click the ‘Add new’ item at the top of each list or in the main menu. This is also
the way to introduce new wind farms and new boundary layer descriptions.
Previously, in Fuga 1, you selected the wind farm layout and turbine type independently. To allow for complex wind
farms this information is now read from the WAsP workspace file. Fugabatch is still able to read Fuga 1 styled input,
but then the calculations are limited to wind farms with one turbine type only. Read more on differences between
Fuga 2 and Fuga 1
Results in the site-list and wind-farm-wake views depend on a reference wind
Workspace data
Wind farm information is stored in WAsP workspace files. Click the ‘Add new’ item at the top of the list to add file or
select one of the previously load workspaces stored in the fuga subdirectory 'WAsP workspaces'.
A workspace can hold several wind farms, which can be organised as nested groups or adjacent independent
projects. Select a turbine group, turbine type and reference site by mouse clicks. Selected objects will be marked by
bold text.
It is possible to temporarily deactivate a turbine site or site group by right-clicking on the object in the workspace
hierarchy. The turbine group called 'Nysted Windfarm' is deactivated in the above image to the right. With only two
groups in the project, a similar selection could have been made by clicking on the 'Rødsand' turbine group.
How to prepare a WAsP workspace
Atmospheric boundary layer data
Select previously calculated wake data or choose ‘Add new’ in the case drop-down list. The names reflect the surface
roughness Z0, inversion height Zi and stability parameter Zeta0, which is surface roughness length divided by the
Monin-Obukhov parameter L.
A new dialog box will pop up if you choosed 'Add new' in the initial drop-down list. In that dialog you select a com-
bination of surface roughness Z0, inversion height Zi and stability parameter Zeta0, which defines that boundary
layer. The parameter defines LUT files, which will be calculated by l preludium.exefor all turbines in the current
project. You can further limit the range of levels, where wake deficit (DAT files) are calculated by trafalgar.exe. Sec-
tion roughness and stability parameters suggests for surface roughness and stability. Typical inversion heights for
neutral atmospheric stability over the North and are something like 400 m. Inversion heights for stable conditions
may be much lower, but bear in mind that Fuga needs a inversion heights higher than the upper tip height of the tall-
est turbine in a project.
The stability parameter Zeta is specified by selecting a value from yet another drop-down list, also included an 'Add
new' option. The values in the list corresponds to existing preLUT directories.
Selecting 'Add new' from the list of stability parameters will invoke a second pop-up menu. This includes a table show-
ing the categorization of Ameya et al. (2011) Wind Energy, Vol. 14, p.767-780. You should not select Zeta0 values
outside this range.
Reference wind
The flow field is defined by a speed and direction at the height of the selected reference site which you select by a
click on the object hierarchy. This can be a turbine site or you could add a reference mast with a different height, say
10-m in the WAsP workspace. Wind at other turbine sites will be corrected by the vertical wind profile only.
Atmospheric boundary layer statistics
single-wake, windfam-wake and other results for a specific wind are calculated for the selected case.
The annual energy production may,optionally, be calculated by a specific boundary layer case applicable at all times
or by a statistically weighted average of multiple boundary-layer cases. When the 'production' tab is activated you will
see this option.
Pressing 'define statistical variation ' will open a file prompt. The present file format is like this but may change in
future version. Having defined a statistical variation you inspect it by selecting 'ABL case Plot' in the 'Production' view.
Result views
Result lists and plots may be exported from the main menu.
Fuga display results in tabulated views on the right-hand side of the program window.
l File overview
l Turbine site list
l Production
l Single wake
l Windfarm wake
A view will be empty if necessary input is missing.
File overview
The files view provides an overview of the data structure. When the structure grows too large it is recommended to
reduce the system by deleting unused subdirectories in the LUTs directory.
It is possible to display input parameters by clicking on turbine and cases. Starting from version 2.2 it is also possible
to display log files containing redirected output whcih in earlier versions was displayed in pop-up DOS windows.
Site list view
The turbine site list displays site location, local wind speed with and without wake effects, electrical power, turbine
efficiency and turbine type. The local wind speed is based on an assumption of a logarithmic wind profile, effects of
orographic or non-uniform surface roughness are ignored in the wake modelling. The power corresponds to the local
wind speed corrected for wake effects. Turbine efficiency is defined as the ratio of actual power and the power cor-
responding to free wind speed.
Before wake calculations the turbine sites are sorted after the wind direction and upwind sites are evaluated before
downwind sites. In this way it is possible to use corrected wind speeds for looking up thrust coefficients at individual
turbine sites.
The steady uniform wind direction assumed in the model is rare in nature. Fuga can average the results over a range
of directions or use its wake meander model. Set a mark in the checkbox above the list to display current averaging
options.
By default the sub-group results include effects of all turbines. Select an individual group in the object hierarchy to
exclude wakes from other turbines.
Production view
This list shows net annual power productions, i.e. the expected production in a typical year corrected for wake
losses. Efficiencies are defined by ratios of net productions and ideal productions without wake effects. Accumulated
results are shown at the bottom of the list, both for all turbines and for sub groups. It is possible to show results for
individual sectors and to compare with the WAsP workspace results.
The Fuga AEP calculation is based on probability-weighted integrals of power productions for all wind speeds and
directions using local wind climates calculated by. Effects of orographic speed up and internal boundary layers are
included in the WAsP generated wind climates. It is just the wake corrections which ignore effects of variable terrain.
This approach is also used in similar WAsP AEP calculations.
By default the sub-group results include effects of all turbines in the set up. Select an individual group in the object
hierarchy to exclude wakes from other turbines.
Single-turbine wake loss plot

A mark in the 'wake loss plot' check box makes Fuga display an overview of the loss of power production due to wake
at individual turbine sites. The colours in this plot indicates percentiles shown in the below report. Use the drop-
down list below the report to select results for all or fo individual wind sectors. Turbine ID and wake loss will be
written on the status bar at the bottom of the Fuga window when the mouse is moved over a turbine position.
ABL case distribution plot
When the production is calculated by a variation of boundary-layer cases, this distribution can be displayed. This is
shown as stacked bars, for variable wind speed or wind direction. Use the radio buttons to the left hand side to select
all directions or certain sub-sectors.
Single wake view
The single wake window shows the wake of a single wake at the selected wind speed. Local wind speeds are shown at
the status bar at the bottom of the program window when the mouse cursor is moved over the wake display. Clicking
on the wake field will change the position of the below cross profile, which is also indicated by a dotted line in the
wake field display. A more precise positioning of the profile is done by the edit boxes left of the profile view. You can
zoom closer to the turbine position and optionally interpolate the results.
When an edit boxes is focused, the profile position may also controlled by keyboard arrow keys combined with the
shift or ctrl key.
Clicking on the radio buttons to the left will change the direction of the chosen transect profile. Use the list below to
select a level for horizontal profiles. The check boxes at the lower-left corner controls whether to normalise distances
in the X-Y plot and whether to plot the wind profile or the wake velocity deficit. Automatic plot scaling is controlled
with the plot options.
The plot scaling is automatic by default. You may also specify axes minima and maxima, this is useful when com-
paring cross profiles at different downwind positions.
Windfarm wake view
.
The wind farm wake window shows the wind field in the entire wind farm at the selected wind speed and direction.
The averaging options, displayed by a mark en the checkbox at the top, will only affect the efficiency for direction
plot.
Zoom and pan
It is possible to zoom in or out with the spin edit box above the plot. The zoom factor is calculated as 2i/5 where i is the
value in the spin edit box. Sometimes it is best to edit this number directly as it takes a while to calculate the plot and
this is done each time the zoom factor change. An alternative zoom method is to press the ctrl key while scrolling the
mouse wheel. Dragging the mouse with the ctrl and left mouse keys pressed will pan the wind-farm view.
Speed up options
Plotting the combined wind field of a big wind farm can be slow. The 'WF plot by FFT' option, i.e. acceleration by fast
Fourier transforms, will speed up the plotting of a big project but not of a small one. It should not be used it in
combination with positive zoom, as wakes from turbines outside the view becomes invisible. Use the 'Block WF plot'
option to temporarily prevent any wind-farm plotting.
Adjust the view
The size of the plot depends on the main window size. Plotting is time consuming because the flow has to be eval-
uated everywhere. Adjusting the resolution, shown in number of pixels, will speed up the process. Setting a res-
olution of zero will imply that Fuga interpolate in the single-wake results even for plotting - this makes most sense
with a high zoom factor Checkboxes are used to add speed- and length scales plus coastline to the plot. It is also pos-
sible to adjust the colour scheme. 'New defaults' are recommended for projector displays and 'X-ray' is best for B&W
printer output.
Measure a distance
The approximate distance and direction between two sites can be measured by dragging with the left mouse-key
pressed. Press shift on the keyboard while releasing the mouse key to keep the measure on the screen.
Select group, turbine site or wind profile reference point
There are two ways to select a turbine site group or individual site
1. by selecting a site in the drop-down list in the 'efficiency for direc-
tion' pane
2. by clicking on the map above. Setting a checkmark at 'snap to tur-
bine position' will move the reference point to the nearby turbine site
Use the drop-down list in the top-left corner to select the height above the
surface. Click the button called 'accurate position' to gain full control of the
reference point, e.g. to select a point directly downwind a specific turbine
Efficiency plot
The chart below the main plot shows the power efficiency or local speed reduction as function of wind direction.
These values are normalized by the undisturbed wind speed and production of a turbine without wake effects. The
direction is indicated to the left of the plot when the cursor is moved over the plot.
Horizontal wind profile

The wind profile along a transect - with downwind, crosswind or skew alignment - can be viewed in a separate chart.
The centre position is selected by clicking on the wind-farm wake plot. Use the checkbox above the main plot to snap
to a nearby turbine position. Having selected a centre position you select the alignment of the transect with the radio
buttons left of the chart and adjust the extend of the transect with the spin edit buttons. Further options are
l Normalised distance - scale the distance with the rotor diameter
l 1D profile - show the profile along the transect at the select level as a curve
l 2D profile - show the 2D profile in a vertical plane along the transect as a coloured map
l Deficit only - show the wake velocity deficit instead of the corrected wind speed
Vertical wind profile
This pane shows the vertical wind profile corrected for the combined wake effect at a position selected by a click on
the map or more in the 'accurate position' dialog.
Further information
l File structure
l Fuga operations at program start up
l Differences between Fuga 2 and Fuga 1
l Averaging options
l How to prepare a WAsP workspace
l How to prepare a WAsP windfarm file
l Case statistics file
l Roughness and stability parameters
l How to select a new turbine site group
l Data export
l Result table
l Main menu
l Fuga scripting
l Fuga speed and memory usage
l Software updates
l References
File structure
The system is installed by unzipping the distribution file.
The root directory contains:

l The Fuga directory includes fuga.exe and associated files plus subdirectories with WAsP data.
l The LUTs directory including lookup tables in the structure explained in a previous memo. Initially this direc-
tory is empty, but it will gradually expand as new cases are added. It is safe to remove part of the LUT file
structure from the disc when cases are no longer needed. Fuga will direct single-wake results (*.dat files)
generated by Trafalgar.exe and the associated *.par option files to the LUTs subdirectories.
TIP: You can change the LUT directory path by editing the fuga.ini file in the Fuga directory, e.g. to a part of
your harddisk with no backup. These voluminous files can be reconstructed at any time.
l The Preludium directory includes preludium.exe, trafalgar.exe and associated files. The input file for
generating the preLUT table, i.e. script1.txt, is not modified by Fuga, but input files for setting up a new tur-
bines and boundary-layer cases, i.e. script2fuga.txt and script3fuga.txt, are redefined according to the
Fuga user choices.
Extended file structure with data for variable atmospheric stability
Starting from Fuga 2.3 we need to store data for variable atmospheric stability. The stability parameter is zeta0,
which is the surface roughness diveded by Monin- Obukhov length. This parameter is built into the LUT and preLUT
file names. We generally have more than one preLUT table and more LUT tables than previously.
Fuga operations at program start up
At the beginning of each session Fuga will scan existing LUTs directories and present available information in drop-
down list on the left-hand side of the program window. It also reads the fuga.ini file in order to recover settings of the
previous session. (This *.ini file is dispensable at any time). At the very first session fuga will not detect any data and
therefore start to calculate the preLUT table. This operation will take about 3 min, but fortunately it is only needed
once.
Differences between Fuga 2 and Fuga 1
Fuga 1
Fuga 1 was designed for homogeneous wind farms with uniform hub heights and identical turbines at all turbine
sites. A system of single-wake velocity fields and look-up tables were stored in a folder structure convenient for the
Preludium and Trafalgar programs. These solutions depend on turbine hub dimension, surface roughness and
boundary layer height. Input turbine power- and thrust coefficient curves were read from a WAsP wind turbine gen-
erator file (*.pow;*.wtg). The wind-farm layout was defined by a WAsP wind farm file (*.wwf), which includes tur-
bine positions and local wind climates needed for accurate annual energy production estimates. Fuga 1 assumed all
turbine hub heights to be identical, so the user had to avoid variable heights in the WAsP project generating the wind
farm file.
Fuga 2
Fuga 2 is designed for complex wind farms and wind-farm clusters with mixtures of turbine types. The Pre-
ludium/Trafalgar data structure is maintained, although we sometimes apply multiple folders for a single turbine if
LUTs and wake data are needed for several hub heights and atmospheric stabilities. The complex wind farm layout is
read from the WAsP workspace file (*.wwh). With this file format it is possible to organize turbine sites into multiple
site groups nested into as many sub groups as needed. Each group, and even the individual turbine site, can have a
specific turbine generator and hub height.
Results are aggregated for each turbine group. Effects of turbines in neighbouring groups are included by default,
but you can also model a turbine group without wake effects from neighbouring groups.
Averaging options
Averaging options affects calculations in
l exported result tables
l turbine site list
l production table (new in Fuga 2.4)
l windfarm-wake 'efficiency for direction' plot (new in Fuga 2.4)
Averaging options does not affects calculations in
l single-wake results
l windfarm-wake main plot
l windfarm-wake horizontal profile
l windfarm-wake vertical profile
One reason for these limitations is that the speed-optimized wake meander model only simulates wake displacements
near downwind turbine positions. The detailed wake trajectory is unknown.
The averaging options appear at the top of the window when applicable. Rotor-averaged wind speed means that the
inflow velocity is evaluated at seven points on the rotor and a weighted average is used for evaluating thrust and
power. Without this option the inflow is evaluated at the rotor centre only. The option can be used in combination with
either directional averaging method.
Directional averaging
Fixed wind direction is the basic and fastest method. It may however overestimate the wake centre-line velocity deficit
and underestimate the width of the wake-affected region.
The simple average and Gaussian average methods apply filters on the wind direction, which reduces the wake veloc-
ity deficit and increases the width of individual wakes. The width or standard deviation of these filters is specified in
the input box appearing to the right of the filer-type option. Fuga actually implements two versions of these methods.
The first one is based on the average of repeated calculations with wind directions distributed around a mean wind
direction. The second version is based on wind directions covering the full 360° horizon, and here the filtering is
done in Fourier space after FFT transforming of the raw results. The first version is used for analysing specific wind
situations and the second one is used in context of AEP estimates, where winds from all directions are always avail-
able. Exported results from the result table dialogue can be calculated by either version.
The final method invokes meander simulation. This is based on a stochastic model and results will differ after each
simulation. To improve the uncertainty of the estimates you can repeat the simulation and use mean results. The opti-
mum number of repeats depends on the size of the windfarm, as random variations from multiple turbines tend to
cancel each other. The results for a single turbine will depend on the number of upwind turbines, so uncertainties for
individual turbines will typically vary with wind direction. The same is often true for an wind farm covering an elon-
gated area. Use the drop-down list at the top to select which simulation to display in tables and efficiency plot. The
button to the right of the repeat number selector can be used to repeat the simulations.
A checkmark at wind direction trend will activate an additional directional averaging filter, which models the effects of
variable wind direction during the sample period. The option should be set when comparing Fuga model predictions
to observed turbine performance binned after an observed wind direction. In most cases this wind direction is meas-
ured at a reference mast and not at the individual turbine sites, and in this case it will be relevant also to select the
spatial decorrelation option modelling the lack of correlation in concurrent wind directions at reference mast and
individual turbine sites. The averaging effect decreases with wind speed and increases with distance between individ-
ual turbine sites and reference site. A project met mast will be an appropriate reference site. Open the workspace in
WAsP to insert a project reference mast and select it by a click in Fuga object hierarchy. The wind direction trend and
spatial decorrelation options are implemented in Fourier space. They are unavailable in the specific-wind situation but
may be calculated from the 'result-table' data export.
How to prepare a WAsP workspace
The work flow is
1. set up a WAsP project, including groups of turbines sites, a wind atlas, and wind turbine generators
2. right-click on a project object and select the ‘do all feasible calculations’ or press F9
3. check that the WAsP calculations were successful, i.e. the wind farm icon in the WAsP object hierarchy
should not include a warning signs but look like this
4. export the workspace file
There is much more detailed information in the WAsP help file, both regarding user interface, modelling and file for-
mats. The section called 'quick start tutorial' is recommended.
Here is a list of basic operations:
Modify a turbine power- and trust-coefficient curve
There are two file formats for this. The old *.pow format is easiest to edit with an ASCII editor. The *.wtg format can
be edited with the WAsP turbine editor. The latter file format may contain several performance tables applicable for
different air density.
Specify a turbine type
You may (but do not have to) specify a turbine type for each turbine site. WAsP will look for a turbine assigned to the
turbine group if site information is missing. If this information is missing too, then WAsP will search for a turbine type
assigned to its parent group or project. The principle is to look for local information before global one.
Modify a turbine height
A turbine hub height in WAsP may differ from its default value in the power-curve file. Right-click on a site or turbine
group and specify the real-world hub height.
Add or modify turbine positions
The interactive way is to right-click on an existing site to edit its coordinates or right-click on a turbine site group to
add more sites. To edit lots of positions it is more convenient to import a site list which can be prepared by an ASCII
editor. The format is described in the WAsP help file. Tip: Export a list from an existing group and use this as a tem-
plate.
Add a reference mast
You can insert a reference mast for a turbine site group. In WAsP it is used as a reference for a windfarm power
curve. In Fuga you can use it as a reference point for specifying wind condition at a height different from the turbine
hub height(s).
Reorganize turbine site groups
You can insert new turbine site groups in a project and add nested sub groups. You can drag turbine sites from one
site group to another.
Add wind climate information
Climate information is either added
l as a wind atlas file representative for the region or
l as locally observed wind statistics.
Observed wind statistics are read from a file, which can be generated from time series with a tool called WAsP Climate
Analyst. It is also possible to write a *.tab file which has a relatively simple format, see WAsP help file. WAsP observed
wind file is assigned to a met station which specifies the met mast position and the station is assigned to a wind atlas.
Wind atlas information are cleansed for local effects.
The number of wind sectors used in both WAsP and Fuga depends on the number of sectors in the input wind cli-
mate. The default is to use 12 sectors. Go back to the WAsP Climate Analyst if you need to change this.
Prepare terrain data
This is normally done with the WAsP Map Editor. For offshore projects the roughness change at the coast line is often
much more significant than terrain elevation on nearby land.
How to prepare a WAsP windfarm file
Fuga 1 needed data in the WAsP windfarm file (*.wwf). Fuga 2 uses the more complex workspace file (*.wwh) but
you can still do Fugabatch calculations based on wind farm files for simple projects with identical turbines.
This wind farm file is prepared in the following way:
1. set up a WAsP project, including a group of turbines sites, a wind atlas, and a wind turbine generator
2. right-click on the site group and select the ‘do all feasible calculations’ option from the popup menu
3. check that the WAsP calculations were successful, i.e. the wind farm icon in the WAsP object hierarchy
should not include a warning signs but look like this
4. right-click on the site group again and select ‘export wind farm to file’ from the popup menu
The WAsP wind farm file contains turbine locations and, with successful WAsP calculations, also the local wind cli-
mates and annual power production estimates.
Fuga version 1.0 used the even simpler WAsP site location *.txt or *.wsg file format. This file type was, however,
abandoned in version 1.1, because it lacks information on wind climate and power production.
Case statistics file
The influence of variable statistics is simplified to a statistical weighting of a selection of specific cases. Each case is
specified by surface roughness z0, inversion height zi and stability parameter zeta0 and may be assigned a text iden-
tifier. The statistical frequency of occurrence for each case is given in sectors and wind-speed bins.
The sectors are defined by their dividing directions, e.g. 348-23°, 23-46°, 46-217°, 217-348° , 23-46°in the below
example. Wind-speed bins are defined by their maxima, e.g. 0-2 m/s, 2-4 m/s, 4-6 m/s, 6-8 m/s, 8-10 m/s, 10-15
m/s, 15-30 m/s in the below example.
Example
The following sample file 'casestats.txt' is included is found in the Fuga directory.
Roughness and stability parameters
The tables provides surface roughness and stability parameters for Fuga. They describe an equilibrium offshore sur-
faces layer with surface roughness estimated by Charnock's equation. The wind speed U [m/s] is defined at a ref-
erence height of 70 m. The inverse Monin-Obukhov length 1/L values [m-1 ]correspond to Sathe-Gryning-Peña class
limits and midpoints.
Table 1: Z [mm] as function of wind speed and stability.
0
very unstable near-neu- neutral near-neu- stable
unstable tral tral
unstable stable
1/L
U -0.0200 -0.0150 -0.0100 -0.0075 -0.0050 -0.0035 -0.0020 0.0000 0.0020 0.0035 0.0050 0.0125 0.0
4.0 0.01624 0.01582 0.01529 0.01496 0.01455 0.01425 0.01386 0.01303 0.01178 0.01096 0.01023 0.00750 0.00
5.0 0.02740 0.02666 0.02572 0.02514 0.02443 0.02390 0.02323 0.02177 0.01961 0.01820 0.01695 0.01231 0.00
6.0 0.04213 0.04094 0.03945 0.03853 0.03740 0.03656 0.03549 0.03320 0.02980 0.02759 0.02564 0.01847 0.01
7.0 0.06073 0.05895 0.05675 0.05538 0.05370 0.05245 0.05088 0.04749 0.04250 0.03927 0.03642 0.02605 0.01
8.0 0.08350 0.08098 0.07786 0.07593 0.07357 0.07181 0.06959 0.06484 0.05786 0.05337 0.04941 0.03510 0.02
9.0 0.11073 0.10730 0.10305 0.10043 0.09722 0.09483 0.09184 0.08542 0.07603 0.07001 0.06471 0.04569 0.03
10.0 0.14269 0.13816 0.13256 0.12909 0.12487 0.12174 0.11781 0.10939 0.09713 0.08930 0.08242 0.05787 0.04
11.0 0.17966 0.17381 0.16661 0.16216 0.15674 0.15272 0.14768 0.13692 0.12130 0.11135 0.10263 0.07168 0.05
12.0 0.22190 0.21452 0.20543 0.19983 0.19302 0.18797 0.18165 0.16817 0.14865 0.13626 0.12543 0.08718 0.06
13.0 0.26968 0.26052 0.24927 0.24234 0.23391 0.22768 0.21988 0.20328 0.17931 0.16414 0.15091 0.10440 0.07
14.0 0.32326 0.31207 0.29833 0.28988 0.27962 0.27203 0.26255 0.24240 0.21339 0.19509 0.17915 0.12338 0.09
15.0 0.38291 0.36941 0.35284 0.34267 0.33033 0.32122 0.30983 0.28569 0.25101 0.22919 0.21023 0.14417 0.10
16.0 0.44888 0.43277 0.41303 0.40092 0.38623 0.37541 0.36189 0.33327 0.29228 0.26654 0.24423 0.16681 0.12
17.0 0.52143 0.50239 0.47910 0.46482 0.44753 0.43479 0.41890 0.38530 0.33730 0.30724 0.28122 0.19133 0.13
18.0 0.60083 0.57853 0.55127 0.53459 0.51440 0.49954 0.48101 0.44191 0.38618 0.35137 0.32129 0.21777 0.15
19.0 0.68734 0.66141 0.62977 0.61042 0.58703 0.56983 0.54840 0.50324 0.43903 0.39902 0.36450 0.24617 0.17
20.0 0.78120 0.75129 0.71481 0.69253 0.66562 0.64584 0.62123 0.56943 0.49595 0.45027 0.41093 0.27655 0.20
21.0 0.88270 0.84839 0.80661 0.78111 0.75034 0.72774 0.69966 0.64061 0.55705 0.50521 0.46065 0.30895 0.22
22.0 0.99208 0.95296 0.90537 0.87636 0.84138 0.81572 0.78384 0.71691 0.62242 0.56393 0.51372 0.34341 0.24
23.0 1.10961 1.06524 1.01133 0.97849 0.93893 0.90993 0.87394 0.79848 0.69216 0.62649 0.57022 0.37996 0.27
24.0 1.23556 1.18547 1.12469 1.08770 1.04318 1.01057 0.97012 0.88543 0.76637 0.69299 0.63021 0.41862 0.30
25.0 1.37020 1.31391 1.24567 1.20419 1.15430 1.11779 1.07253 0.97790 0.84516 0.76351 0.69375 0.45942 0.32
Table 2: 106 ×Zeta as function of wind speed and stability.

0
very unstable near-neu- neutral near-neu- stable
unstable tral tral
unstable stable
1/L
U -0.0200 -0.0150 -0.0100 -0.0075 -0.0050 -0.0035 -0.0020 0.0000 0.0020 0.0035 0.0050 0.0125 0.0
4.0 -0.3248 -0.2373 -0.1529 -0.1122 -0.0728 -0.0499 -0.0277 0.0000 0.0236 0.0384 0.0511 0.0937 0.1
5.0 -0.5480 -0.3998 -0.2572 -0.1886 -0.1222 -0.0837 -0.0465 0.0000 0.0392 0.0637 0.0847 0.1538 0.1
6.0 -0.8426 -0.6141 -0.3945 -0.2890 -0.1870 -0.1279 -0.0710 0.0000 0.0596 0.0966 0.1282 0.2308 0.2
7.0 -1.2146 -0.8843 -0.5675 -0.4153 -0.2685 -0.1836 -0.1018 0.0000 0.0850 0.1375 0.1821 0.3256 0.3
8.0 -1.6700 -1.2148 -0.7786 -0.5695 -0.3678 -0.2513 -0.1392 0.0000 0.1157 0.1868 0.2471 0.4388 0.5
9.0 -2.2146 -1.6095 -1.0305 -0.7532 -0.4861 -0.3319 -0.1837 0.0000 0.1521 0.2450 0.3236 0.5712 0.6
10.0 -2.8538 -2.0723 -1.3256 -0.9682 -0.6244 -0.4261 -0.2356 0.0000 0.1943 0.3125 0.4121 0.7234 0.8
11.0 -3.5931 -2.6072 -1.6661 -1.2162 -0.7837 -0.5345 -0.2954 0.0000 0.2426 0.3897 0.5131 0.8960 1.0
12.0 -4.4379 -3.2178 -2.0543 -1.4988 -0.9651 -0.6579 -0.3633 0.0000 0.2973 0.4769 0.6271 1.0897 1.2
13.0 -5.3935 -3.9079 -2.4927 -1.8175 -1.1696 -0.7969 -0.4398 0.0000 0.3586 0.5745 0.7545 1.3050 1.5
14.0 -6.4652 -4.6811 -2.9833 -2.1741 -1.3981 -0.9521 -0.5251 0.0000 0.4268 0.6828 0.8957 1.5423 1.8
15.0 -7.6581 -5.5411 -3.5284 -2.5700 -1.6516 -1.1243 -0.6197 0.0000 0.5020 0.8022 1.0511 1.8022 2.1
16.0 -8.9776 -6.4915 -4.1303 -3.0069 -1.9312 -1.3139 -0.7238 0.0000 0.5846 0.9329 1.2211 2.0851 2.4
17.0 -10.4287 -7.5359 -4.7910 -3.4862 -2.2376 -1.5218 -0.8378 0.0000 0.6746 1.0753 1.4061 2.3917 2.7
18.0 -12.0166 -8.6779 -5.5127 -4.0094 -2.5720 -1.7484 -0.9620 0.0000 0.7724 1.2298 1.6065 2.7222 3.1
19.0 -13.7467 -9.9212 -6.2977 -4.5782 -2.9352 -1.9944 -1.0968 0.0000 0.8781 1.3966 1.8225 3.0771 3.5
20.0 -15.6241 -11.2693 -7.1481 -5.1940 -3.3281 -2.2604 -1.2425 0.0000 0.9919 1.5760 2.0547 3.4569 4.0
21.0 -17.6539 -12.7258 -8.0661 -5.8583 -3.7517 -2.5471 -1.3993 0.0000 1.1141 1.7682 2.3032 3.8619 4.4
22.0 -19.8416 -14.2944 -9.0537 -6.5727 -4.2069 -2.8550 -1.5677 0.0000 1.2448 1.9737 2.5686 4.2926 4.9
23.0 -22.1922 -15.9786 -10.1133 -7.3387 -4.6947 -3.1848 -1.7479 0.0000 1.3843 2.1927 2.8511 4.7494 5.4
24.0 -24.7113 -17.7821 -11.2469 -8.1577 -5.2159 -3.5370 -1.9402 0.0000 1.5327 2.4255 3.1510 5.2327 6.0
25.0 -27.4040 -19.7086 -12.4567 -9.0314 -5.7715 -3.9123 -2.1451 0.0000 1.6903 2.6723 3.4688 5.7428 6.5
How to select a new turbine site group
Aggregated results shown for subgroups of turbine sites normally includes wake effects of turbines in neighbouring
groups. If you want to see the performance of a group without the effects of neighbouring groups you can select the
group of interest in the object hierarchy.
Working with projects with three or more turbine site groups you may want to include all but one or some of these
groups. This can be done by first selecting the combined set of turbines and then temporarily deactivating sub-
groups. You do this by right-clicking on a group.
Data export
Data export to file or clipboard are found in the 'file' menu. Clicking on one of the 'Save...' menu items will invoke a
Windows save-file dialogue. The default names are shown below but it is also possible to specify other names. Most of
the results are identical to the ones shown in the user interface, but the result table has a more flexible layout. The
'Zip all' option will collect all results in a single file. The 'Copy...' options will copy results to the Windows clipboard.
Result table
The layout of exported result tables can be modified in various ways
l Result type -chose whether to export wind speeds at turbine locations, with wake effects from neighbour tur-
bines, or the corresponding power read from the WAsP power curve file.
l Turbine site - select data from a single site, from all sites, or the average or total from the entire windfarm
l Wind direction - select a range of wind directions and a increment. Choose output for each direction of just
the average
l Wind speed - select a range of wind speeds and a increment.
The results will be modelled with current averaging options.
Main menu
You can new projects and case data form the File menu. It is possible to export results and plots to file and clipboard.
The about box under the help menu provides version information for the Preludium, Trafalgar and Fuga programs.
Fuga scripting
FugaBatch is a special version of Fuga designed to run in a non-interactive mode. The calculations are controlled by a
script input file and the name of this file is specified as a parameter in the command line, e.g.
fugaBatch fugaScript.txt [/L=MyLogFile.txt]
The second parameter is optional. When included, the progress reports normally written to a DOS window are redi-
rected to the specified file.
The script input file has two sections. First the setup is specified by names of various input files. These file names are
written with double quotation marks and must be given in the order shown the below examples, since Fuga will proc-
esses each line of the script individually. The easiest way to generate the input files is by using Fuga in interactive
mode, but you can also generate them by calling WAsP, preludium and trafalgar directly. It is not possible to change
the Fuga setup within a single call, i.e. if you want to model a wind farm with two alternative turbine types in non-
interactive mode you have to call fugaBatch twice.
Setup
The first section can be written in two ways:
Fuga 1 style (Deprecated)
Command: SiteListFilename
the full path to a WAsP windfarm file (*.wwf)
Command: TurbineFilename
the full path to a WAsP turbine file (*.pow or *.wtg)
Command: CaseFilename
the full path to the casedata file (casedata.bin) in a directory with Preludium and Trafalgar output.
NB: You may need to update old scripts as Fuga 2.3 GUI changes names of existing directories, e.g. from "C:\LUTs-
T\Vestas_V80_(2_MW_offshore)[h=67.00]\Z0=0.00010000Zi=00399\casedata.bin" to "C:\LUTs-T\Vestas_V80_
(2_MW_offshore)[h=67.00]\Z0=0.00010000Zi=00399Zeta0=5.0E-7\casedata.bin"
Fuga 2 style
Command: Workspace
the full path to a WAsP workspace file (*.wwh)
Command: CASE z0 zi zLo zHi [Zeta0]
Boundary layer specifications. Here, z0 is surface roughness [m], zi is the boundary layer height [m], and zL0 and
zHi [m] define the vertical range of wake output. The zi, zLo, and zHi parameters are approximate as they will be
rounded off to the nearest level in the preludium model. The stability parameter Zeta0, which is z0 divided by Moni-
Obukhov length L, is optional. Stability will be set to neutral, Zeta0=0, if not specified.
Modified Fuga 1 style
Command: SiteListFilename
Command: TurbineFilename
Command: CASE z0 zi zLo zHi [Zeta0]
The original Fuga 1 style script normally only works with input files created by the main Fuga program. Scripts with
the CaseFilename substituted by the newer CASE command make Fugabatch generate missing information.
Instructions
The second section of the script input file contains instructions for various kinds of output. You can include as many
instructions as you like in a single script.
Annual energy production
Command: AEP Nsec [caseStatsFilename] filename
where Nsec actually now is redundant. It is only allowed for backward compatibility and it will be ignored if present.
Filename is the name of the output file. CaseStatsFilename is optional. If omitted, the calculations will be based on
boundary-layer conditions in the 'caseFilename' (Fuga 1 style) or 'case' (Fuga 2 style) instructions. When included,
the AEP calculations are based on frequencies of different atmospheric stabilities defined in the specified file . The file
format is described here.
Hub-height wind speeds at turbine positions
Command: SHELTER filename
where filename is the name of the output file This will produce a table with windspeeds at turbine positions for all tur-
bine sites, all directions and all free stream wind speeds. The calculated wind speeds include wake effects of all
upwind turbines, but not the effect of the local turbine
Reference site
Command: REFERENCESITE x-coordinate y-coordinate height
Where x-coordinate, y-coordinate and height gives the reference site position. The horizontal coordinates position are
used when DirStatFilter is set to 2 in the AVERAGINGOPTIONS command. The height is used as the reference height
for wind speeds defined in the PROD, FLOW and TRANSECT commands.
Power production and local hub-height wind speed for a given wind situation
Command: PROD windspeed windDirection filename
Where windspeed and windDirection describes the free wind and filename is the name of the output file
Hub-height wind speeds in a cartesian grid
Command: FLOW windspeed windDirection X0 Y0 dX dY Nx Ny filename
Where windspeed and windDirection describes the free wind, X0 and Y0 are the coordinates of the lower-left corner
of the grid, dX and dY are grid resolutions in X and Y direction, Nx and NY are the number of grid nodes in each direc-
tion and filename is the name of the output file. The output is the wind affected by wakes of all turbines, including the
local turbine if a grid node matches a turbine site. Thus the results are different than those produced by the SHELTER
command.
The output file is in the Surfer Ascii *.grd format described in the WAsP help file.
Wind speeds in a 2D cartesian grid intersecting a 3D wake
Command: FLOW windspeed windDirection X0 Y0 Z0 rot tilt dX dY Nx Ny filename
where windspeed and windDirection describes the free wind, X0, Y0, and Z0 are coordinates of the lower-left corner
of the grid, as in the previous version. Furthermore, dX and dY are grid resolutions and Nx and NY are the number of
grid nodes in each direction and filename is the name of the output file. An important change is, however, that the
Surfer grid coordinate system does not have to be oriented in a horizontal plane with with X toward East. The grid
orientation is specified by 1) a counter-clockwise rotation angle, e.g. to make the surfer grid X axis point NW you
enter rot=135, and 2) a tilt angle, e.g. to make the Y axis point vertical you enter tilt=90. The grid origo is centred at
the point (X0, Y0, Z0). The wind speed applies to the turbine hub height.
Hub-height wind speeds along a transect
Command: TRANSECT windspeed windDirection X1 Y1 X2 Y2 N filename
Where windspeed and windDirection describes the free wind, X1 and Y1 are coordinates of the start of the transect,
X2 and Y2 are coordinates of the end of the transect, N is the number of nodes along the transect and filename is the
name of the output file. These results are similar to those found by the FLOW command.
Wind speeds along a general 1D transect in a 3D wake
Command: TRANSECT windspeed windDirection X1 Y1 Z1 X2 Y2 Z2 N filename
As above, except that the now the transect start and end points are written with three coordinates. The wind speed
applies to the turbine hub height.
Velocity deficit
The DEFICITcommand is similar to the two versions of the TRANSECTcommand. The only difference that the velocity def-
icit is written to the file instead of the wake-corrected wind field.
Result Table
Command: RESULTTABLE ResultType TurbineSite uLo uHi Du DirLo DirHi Ddir DirAve useFFT filename
Where
l ResultType is set to 0 for wind speed or 1 for power production
l TurbineSite is a string which either is identifier of a particular turbine site or "All" for data from all sites or
"Total" for the average(speed) or aggregated(power) of for all sites in the wind farm
l uLo is the lower limit of the wind-speed range
l uHi is the upper limit of the wind-speed range
l dU is an wind-speed increment in the table
l DirLo is the first wind direction
l DirHi is the last value in a range of wind direction which is arranged anticlockwise
l Ddir is a wind-direction increment in the table
l DirAve indicates whether to output for every wind direction (if set to 0) or to output the average for all direc-
tions in the range (if set to 1)
l useFFT set to 1 means that results are calculated for all directions and the directional filters are applied in Fou-
rier space. With this option set there may be a small correction to the wind-direction increment Ddir and the
last wind direction DirHi. With useFFT set to 0 (default) the top-hat and Gaussian averaging filters will be cal-
culated as the weighted average of several calculations, and the DirStatFilter corrections defined in the AVER-
AGINGOPTIONS command are temporarily inactive.
l Filename is the name of the output file
Averaging options
Command: AVERAGINGOPTIONS RotorAve AveFilter FilterScale MeanderRepeats DirStatFilter
Where
l RotorAve is set to 1 for rotor-averaged inflow velocity or 0 for rotor centre velocity (default)
l AveFilter is set to
o 0 for a fixed direction with no directional averaging (default)
o 1 for simple top-hat averaging in a user-defied sector
o 2 for weighted averaging in a Gaussian distribution around the centre angle
o 3 for meander simulation
l FilterScale is the half width of the top-hat averaging filter or spread of the Gaussian filter. This parameter has
no effect on the meander model, but please insert a dummy value anyway
l MeanderRepeats is the number of repeated meander simulations.
l DirStatFilter models the statistical uncertainty of power statistics due to variable wind direction. It is set to
o 0 for no correction
o 1 for correction for wind-direction trend during each sample period (typically a 10-min sample period)
o 2 for correction of trend effects as above, plus statistical de-correlation due to spatial separation
between reference mast and turbine sites. NB: Remember to set the REFERENCESITE command for
this option
Averaging options affect subsequent script commands until a new set of options is specified. Averaging options will
only affect AEP, PROD, SHELTER and RESULTTABLE commands - and with further exceptions for the DirStatFilter
option. This last option relies on FFT calculus and therefore it is not active for the PROD calculations and for RESUL-
TABLE it is only active with the useFFT option set to 1. It is also inactive if AveFilter option is set to 0 (fixed direction).
If that case is of interest you could use a top-hat filter of zero width. Read about equivalent averaging options in the
GUI version.
Pause in output
Command: PAUSE [duration]
Where
l Duration is the length of the pause in seconds. With no duration specified, the script will wait for the user to
press 'Enter' on the keyboard.
This command will insert a delay in the screen output. It is intended for script testing and will be ignored if the output
is redirected to a log file.
Examples
A Fuga 1 styled script file
[Setup]
sitelistFilename ".\WAsP sitelists\Horns Rev 1.wwf"
turbineFilename ".\WAsP turbine data\Vestas V80 (2000 kW).pow"
caseFilename "..\LUTs\Vestas_V80_(2_MW_offshore)\Z0=0.00010000Zi=00400Zeta0=0.00E-0\casedata.bin"
CASE 0.0001 400 1 400
[Instructions]
AEP 12 ".\export\Horns Rev AEP.txt"
SHELTER ".\export\Horns Rev shelter factors.txt"
PROD 10 282 ".\export\Horns Rev u=10 Dir=282.txt"
FLOW 10 265 423000 6147000 10 10 1000 600 ".\export\Horns Rev u=11 Dir=270.grd"
FLOW 7 215 429905 6147152 5 144 90 20 2 400 160 ".\export\Horns Rev vertical diagonal cross section, u=10
Dir=265.grd"
TRANSECT 10 270 423000 6147000 433000 6147000 200 ".\export\Horns Rev transect u=10 Dir=270.txt"
TRANSECT 7 215 426703 6149495 5 426703 6149495 350 346 ".\export\Horns Rev vertical transect, u=7
Dir=215.txt"
ResultTable 0 "All" 5 8 1 265 275 1 0 2 4.2 ".\export\Horns Rev results.txt"
A Fuga 2 styled script file
[Setup]
WORKSPACE ".\WAsP workspaces\hornsrev.wwh" "Horns rev combined"
CASE 0.0001 400 1 400
[Instructions]
AEP ".\export\Horns Rev AEP.txt"
SHELTER ".\export\Horns Rev shelter factors.txt"
FLOW 10 265 423000 6147000 10 10 1000 600 ".\export\Horns Rev u=11 Dir=270.grd"
FLOW 7 215 429905 6147152 5 144 90 20 2 400 160 ".\export\Horns Rev u=11 Dir=270.grd"
TRANSECT 10 270 423000 6147000 433000 6147000 200 ".\export\Horns Rev transect u=10 Dir=270.txt"
TRANSECT 7 215 426703 6149495 5 426703 6149495 350 346 ".\export\Horns Rev transect u=7 Dir=215(3).txt"
DEFICIT 10 270 423000 6147000 433000 6147000 200 ".\export\Horns Rev deficit u=10 Dir=270.txt"
DEFICIT 7 215 426703 6149495 5 426703 6149495 350 346 ".\export\Horns Rev deficit u=7 Dir=215(3).txt"
ResultTable 0 "All" 5 8 1 265 275 1 0 2 4.2 ".\export\Horns Rev results.txt"
A Fuga 1 styled script file with stability effects
[Setup]
sitelistFilename ".\WAsP sitelists\Horns Rev 1.wwf"
turbineFilename ".\WAsP turbine data\Vestas V80 (2000 kW).pow"
caseFilename "..\LUTs\Vestas_V80_(2_MW_offshore)\Z0=0.00010000Zi=00400Zeta0=2.00E-7\casedata.bin"
CASE 0.0001 400 1 400 2.00E-7
[Instructions]
AEP "caseStats.txt" ".\export\Horns Rev AEP.txt"
A Fuga 2 styled script file with stability effects
[Setup]
CASE 0.0001 400 1 400 2.00E-7
[Instructions]
AEP "caseStats.txt" ".\export\Horns Rev AEP.txt"
A result table script
[Setup]
CASE 0.0001 400 1 400
[Instructions]
ResultTable 0 "All" 5 8 1 265 275 1 0 ".\export\Nysted results.txt"
AVERAGINGOPTIONS 1 2 3.5 1 2
ResultTable 0 "Total" 5 8 1 0 359 1 0 1 ".\export\Nysted results, RotorAve, Gauss(3.5Deg), decorrelation and
FFT.txt"
A script demonstrating averaging options

[Setup]
CASE 0.0001 400 1 400
[Instructions]
AVERAGINGOPTIONS 0 0 0 1
AEP ".\export\Horns Rev AEP (no averaging).txt"
AEP ".\export\Horns Rev AEP (rotor averaging).txt"
AVERAGINGOPTIONS 0 1 4.0 1
AEP ".\export\Horns Rev AEP (simple directional averaging).txt"
AVERAGINGOPTIONS 0 2 3.2 1
AEP ".\export\Horns Rev AEP (gaussian directional averaging).txt"
AVERAGINGOPTIONS 0 3 0 1 0
AEP ".\export\Horns Rev AEP (meander simulation).txt"
PROD 11 270 ".\export\Horns Rev u=11 Dir=270.txt (meander simulation).txt"
PROD 11 270 ".\export\Horns Rev u=11 Dir=270.txt (mean of ten meander simulations).txt"
AEP ".\export\Horns Rev AEP (meander simulation with trend filter).txt"
REFERENCESITE 435253 6149502 60
AEP ".\export\Horns Rev AEP (meander simulation with de-correlation filter).txt"
PROD 11 270 ".\export\Horns Rev u=11 Dir=270.txt (no averaging, 10m reference height).txt"
PROD 11 270 ".\export\Horns Rev u=11 Dir=270.txt (no averaging, 60m reference height).txt"
PAUSE 5
Exit code
0. Normal
1. Error reading site list
2. Error reading turbine file
3. Error reading case data fil
4. Error reading workspace file
5. Error calculating AEP
6. Error calculation PROD - production for specific wind
7. Error calculating turbine shelter factors
8. Error exporting 2D wind field in Surfer format
9. Error exporting 1D wind transect or deficit
10. Error reading CASE command
11. Error exporting result table
Fuga speed and memory usage
The Fuga user must make some choices, which affect time of computation and memory usage. These choices will be a
compromise between accuracy and computational efficiency.
The number of levels in LUT tables

Fuga calculates 2D solutions for horizontal layers and use vertical interpolation when necessary. The computation
time for creating the lookup tables (LUT) is proportional to the number of levels, which you decide when defining a
boundary-layer case. If you never need to plot vertical velocity profiles, you can save time of computation and disk
space by limiting the number of calculation levels. Without rotor averaging, it is sufficient to define a range including
all turbine hub heights. With rotor averaging, the range must cover all turbine rotors in a project from hub height to
71% of upper and lower tip height.
Number of turbine types

A set of single-wake LUT tables is calculated or reloaded for each turbine type in a project. This information is
removed from memory when selecting a new boundary-layer case or WAsP workspace. Note that Fuga considers a tur-
bine as of a new type when the hub height differs. The memory requirements are proportional to the number of tur-
bine types, but the computation time is not much affected once the single-turbine wake data are available.
Number of turbines
The wind-farm wake calculation time increases with the number of turbines. The calculation of wake effects on
individual turbines will depend on the number of upwind turbines, though not when these are far off the wind tra-
jectory hitting the downwind turbine. Thus, the overall wake calculation time scale between O(N) and O(N2), where N
is the number of turbines.
Rotor averaging
With rotor averaging, the inflow is calculated as a weighted average of the velocity of seven points on the exposed tur-
bine. This extra work slows down the calculations by a factor of approximately 4.
Meander model
The meander model is a stochastic simulation and slower than the simple methods. The wake effect on each turbine is
simulated independently. For AEP calculations we cheat a little and use the same simulated wake trajectories for all
wind speeds, tough with wind-speed dependent thrust. If you select multiple meander simulations, the simulation
time will increase proportionally. A high number of simulations is usually unnecessary for a large wind farm as var-
iable results for individual turbines tend to cancel each other. The variability for individual turbines will generally
decrease with the number of upwind turbines and also disappear for no upwind turbines are found. Thus, variability
of single-turbine meander results will depend on direction.
Number of boundary-layer cases

When calculating the AEP by a mixture of boundary-layer cases, a set of LUTs will be loaded into memory for each
case and for each turbine type. Using cases statistics with many reference cases in combination with multi-turbine-
type projects may cause a memory shortage for win32 versions of Fuga and FugaBatch. The problem is that a win32
program only address up to 1.5 GB RAM. The AEP calculation time for mixed-case statistics is proportional to the
number of boundary-layer cases.
Option for calculation blocking

Automatic updates can be annoying when fine-tuning big Fuga projects, e.g. when you want to deactivate certain tur-
bines or turbine groups. You can avoid automatic updates by setting checkmarks at ‘production| block production
estimates’ and ‘windfarm wake| block plot’.
CPU cache
Fuga calculations tend to speed up the second time in a work session. This is probably because LUT table data are
available in the cache.
Software updates
We recommend to install software updates in a new directory, check that the new system works, and then delete the
old file structure. However, for smaller changes, we can give the following guidelines:
l Changes in preludium.exe or its input parameters in script1.txt will outdate all LUTs subdirectories, which
should be deleted.
l Changes in trafalgar.exe will outdate *.dat files at the the second level of the LUTs subdirectories, i.e. the
case level. If trafalgar.exe is changed it is probably be easiest to delete all case subdirectories and regen-
erate them.
l Changes in fuga.exe should only affect files in the fuga directory and its subdirectories.
l With new turbine information, e.g. a modified power curve, all files in the corresponding turbine directory
and its case directories should be removed.
References
l Søren Ott (2011): Linearized CFD, Risø-I-3093 (EN), 29 pp.
l Søren Ott (2011): Linearized CFD Models for Wakes, Risø-R-1772 (EN), 37 pp.
l Søren Ott and Morten Nielsen (2010): Fuga Files and Scripts
l Ameya Sathe, Sven-Erik Gryning, Alfredo Peña (2011): Comparison of the atmospheric stability and wind pro-
files at two wind farm sites over a long marine fetch in the North Sea, Wind Energy Vol. 14, Issue 6, p. 767–
780
l WAsP help file, distributed with WAsP and also available from www.wasp.dk
Bibliographic Data Sheet DTU Wind Energy E-0046
Title and author(s)
Developments of the offshore wind turbine wake model Fuga
Søren Ott and Morten Nielsen
ISBN ISSN
978-87-92896-81-0
Dept. or group Date
DTU Wind Energy January 22, 2014

Groups own reg. number(s) Project/contract No.
Pages Tables Illustrations References
89 3 21 28
Abstract (Max. 2000 char.)
This is the final report of the project entitled Risø DTU Modelling Services carried out by
DTU Wind Energy (formerly known as Risø National Laboratory) as part of the Carbon
Trust’s Offshore Wind Accelerator Stage 2 under a contract with Carbon Trust. The project
is a follow–up to a Carbon Trust’s Offshore Wind Accelerator Stage 1 project called Lin-
earized CFD Wake models. The earlier project resulted in the development, implementation
and validation of the Fuga model. Fuga is a linearized CFD model that can predict wake
effects for offshore wind farms. The main purpose of Stage 2 is to add more features to Fuga
and turn it into a useful tool for offshore wind farm developers. The new features consist in
• Flexibility. Including the ability to cope with several types of turbines in the same
project, thus making it possible to predict inter farm interactions. The graphical user
interface has been greatly improved and a number of input/output facilities have been
added.
• Stability effects. The effect of stability has been added through a modification of the
eddy viscosity based on Monin–Obukhov theory. The numerical solver developed in
Stage 1 has been generalized in order to make it deal with the modified equations.
• Meandering. Meandering has been included in the form of a post processing of the
model results that bend and twist the wake centreline. The meandering centrelines are
calculated using a Gaussian process developed on the basis of measured spectra. An
analysis of meteorological data from Horns Rev has been made in order to quantify the
impact of non–stationarity of the wind direction. The results are generalized so as to
account for the uncertainties imposed by a ten minute mean value trend as well as by
the distance between turbines and the met mast.
The old model has been validated against a number of data sets. Some of these tests have
been repeated in order to demonstrate and validate the new model features. Production data
from Horns Rev 1 have been re–analysed using well defined selection criteria for which the
developed uncertainty models apply, and a comparison with data is made. Even if the model
predictions fall within estimated error bars, the model seems to over predict the measured
efficiencies by a few percent.
The model works best for unstable, neutral and light stable conditions whereas the results
for stable and very stable conditions are questionable. We suspect this is caused by a failure
of the numerical solver that becomes progressively more severe as the stability increases.
Descriptors
DTU Wind Energy is a department of the Technical University of Denmark with a unique integration of research, education, innovation and
public/private sector consulting in the field of wind energy. Our activities develop new opportunities and technology for the global and Danish
exploitation of wind energy. Research focuses on key technical-scientific fields, which are central for the development, innovation and use of wind
energy and provides the basis for advanced education at the education.
We have more than 240 staff members of which approximately 60 are PhD students. Research is conducted within nine research programmes
organized into three main topics: Wind energy systems, Wind turbine technology and Basics for wind energy.
Technical University of Denmark

Department of Wind Energy
Frederiksborgvej 399
Building 118
4000 Roskilde
Denmark
Phone 46 77 50 85
info@vindenergi.dtu.dk
www.vindenergi.dtu.dk

DTU Wind Energy E 0046

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

DTU Wind Energy E 0046

Uploaded by

Copyright:

Available Formats

Downloaded from orbit.dtu.

dk on: Sep 16, 2020

Developments of the offshore wind turbine wake model Fuga

Ott, Søren; Nielsen, Morten

Link back to DTU Orbit

Søren Ott and Morten Nielsen

DTU Wind Energy E-0046

Approved by: Hans E. Jørgensen

Checked by: Morten Nielsen

3 Atmospheric stability modelling 7

8 Appendix A: The Fuga help system 42

DTU Wind Energy E-0046 3

DTU Wind Energy E-0046 5

6 DTU Wind Energy E-0046

DTU Wind Energy E-0046 7

8 DTU Wind Energy E-0046

φ(ζ) = (1 + a ζ)n (10)

DTU Wind Energy E-0046 9

3.2 Model equations

10 DTU Wind Energy E-0046

DTU Wind Energy E-0046 11

Real world Predictions

12 DTU Wind Energy E-0046

P∆θ (x) = β/2 e−β|x| (26)

Table 1. Constants to be used in (27).

Stability a1 [-] a2 [-] b1 [s/m] b2 [s/m]

DTU Wind Energy E-0046 13

Integrating this we find

14 DTU Wind Energy E-0046

Fuga + 5deg filter

Technical University of Denmark

The stationarity analysis can be extended to include effect of inhomogeneity in a simple

DTU Wind Energy E-0046 15

In terms of math we simply replace (29) by

Again the Fourier transform is relatively simple:

4.2 Turbulent scales

16 DTU Wind Energy E-0046

DTU Wind Energy E-0046 17

Courtney & Troen 1990

18 DTU Wind Energy E-0046

4.3 Wake meandering modelling

DTU Wind Energy E-0046 19

0 20 40 60 80 100 120 140

20 DTU Wind Energy E-0046

hv(t) v(t + τ )i = σu2 fv (σu τ ) (37)

A formal solution that gives (39) can be written as

where N is δ correlated Gaussian white noise, i.e

DTU Wind Energy E-0046 21

22 DTU Wind Energy E-0046

The thrust is usually evaluated from the simple formula

DTU Wind Energy E-0046 23

24 DTU Wind Energy E-0046

6.1 Repeats of old tests

DTU Wind Energy E-0046 25

26 DTU Wind Energy E-0046

Description Class L interval h1/Li−1

DTU Wind Energy E-0046 27

28 DTU Wind Energy E-0046

DTU Wind Energy E-0046 29

30 DTU Wind Energy E-0046

-1.5 ´ 10-6 -1. ´ 10-6 -5. ´ 10-7 0 5. ´ 10-7 1. ´ 10-6

DTU Wind Energy E-0046 31

7.2 Roter averaging

7.3 How to make a climatology with stability