u (t ) = U + u r (t ) + u g (t ) + ut (t ) (1)
0 t 〈Ti
t − Ti
ur (t ) = Ar Ti 〈t 〈Ts (2)
Ts − Ti
Ar t 〉Ts
Figure 1: Sinusoidal and IEC-type gusts included in a
standard wind speed time series with a 4 Hz sampling
This component is not of special interest since it frequency. Kaimal spectrum with: I=0.15; U=10 m/s;
introduces a change in the average wind speed, thus L1=150 m and Nf=100. Sinusoidal type parameters:
limiting the specified properties of a ten minutes Ti=20 s, Ts=30 s, Ag=7 m/s; IEC type parameters:
standard simulation, in which the average wind speed Tc=70 s, σg =20 s, Ag=7 m/s.
should be made constant. However, it can be used to
analyze the turbine controllers’ behaviour and the
generator controllers.
The gust component can have different
implementations. Until recently it was just a 2.1 The Kaimal model
deterministic function with several possible
definitions, being the sinusoidal function one of the The Kaimal spectrum model is, like other ones,
most used ones, [8], [4]. One possible definition is in usually presented under a PSD function. For the
(3), where Ag is the gust amplitude: longitudinal wind speed component, the model
defined by Kaimal relates frequency, length scale and
average wind speed, [1]:
0 t 〈Ti
A
g t − Ti
u g (t ) = ⋅ 1 − cos 2π Ti 〈t 〈Ts (3)
fS ( f )
=
4 fL1 / U
(4)
2 Ts − Ti σ 2
(1 + 6 fL1 / U )5 / 3
0 t 〉Ts
where S(f) is the auto-spectral density function, f is
However, in extreme wind conditions that wind the frequency, σ is the standard deviation, L1 is the
turbines must support, the gusts experienced in real length scale and U the average wind speed, all in the
situations are of stochastic nature and have a limited upwind direction.
spatial extension, not affecting the all wind park, [6]. The length scale is dependent on the surface
Somewhat different is the gust definition roughness, z0, as well as the height above ground, z.
included in the relevant IEC standard, [4]. This gust Standards used for wind turbine loading calculations
type includes an initial wind speed decreasing, thus specify different turbulence spectra and/or different
implying additional robustness in the turbine length scales to be used in different test conditions.
controllers. Figure 1 shows an example of the two The length scale dependence on the height above
gust types included in a one minute wind speed time ground can be simplified until a constant as is done in
series. the Danish standard DS472 or in the IEC standard
The turbulent component can have different 61400-1.
forms and is usually represented by a Power Spectral For the wind speed longitudinal component the
Density function, PSD. The most used ones to model standard deviation σ is approximately constant with
height. The standard deviation depends on the of the tower. For a three-blade turbine, this frequency
turbulence intensity and the average wind speed, and is referred to as 3p.
is given by (5):
I = σ /U (5)
2 + bs
H1 ( s ) =
( )(
2 + bs a 1 + b / a s ) (7)
1
2.5 Wind and torque time series Pa = πρR 2u 3C p (λ, β) (13)
2
A technique is needed to synthesize wind speed
signals from the modelled one-dimensional power The power coefficient Cp(λ, β) is a turbine
spectrum with predefined turbulence flow conditions. specific function defining the turbine ability to
One of the most used is the Shinozuka method, which convert the wind kinetic energy into mechanical
allows generating a one-dimensional wind speed energy. Cp(λ, β) is a nonlinear function of the pitch
signal as a realization of a turbulent process, using the angle, β, and the tip speed ratio, λ.
model power spectrum as a target. Each point in the The tip speed ratio is given by:
time series is computed by summing weighted cosine
series with a random phase angle, as expressed in ωr R
λ= (14)
(10): ueq
Nf
Generally, the power coefficient is adequately
u (t ) = U + ∑ S ( f n )∆f cos(2πf n t + φ n + ∆φ) (10) described by (15), [15], in which coefficients c1 to c7
n =1
are turbine specific calculated:
In (10), S(fn) is the spectral density model, Nf is
c5
the considered discrete frequencies number (usually c −
between 50 and 100), ∆f is the incremental frequency, C p (λ, β) = c1 2 − c3β − c4 e λ i (15)
fn is the frequency, φn its uniformly distributed λi
random phase angle and ∆φ is a random phase needed
to achieve non repetitiveness of each simulation run. 1 1 c
with = − 37 .
The generated synthetic signal is only one of the λ i λ + c6β β + 1
possible realizations of a process with the chosen flow
turbulence characteristics because of its random phase
In variable speed systems, and below nominal
angle. The inverse discrete Fourier transform can also
wind speed, the power coefficient is usually
be used to generate the time series, (11):
maintained at an optimum value in order to maximize
the extracted energy.
Nf N The rotor aerodynamic torque, (16), can easily be
u (t ) = U + ∑ An cos(2πf n t ) + ∑ B n sin( 2πf n t ) (11)
n =1 n =1
determined by dividing the rotor power by the rotor
shaft angular velocity:
Essentially, the random phase angle in (10) is
replaced by zero mean value independent normally 1
Ta = πρR 2u 3C p (λ, β) (16)
distributed random coefficients An and Bn. The 2ωr
variance of the process is maintained by imposing the
correct standard deviation in the random distribution, In the most important wind speed operating range
(12): (Region 2), a maximum Cp implies a constant tip
speed ratio and a fixed pitch angle. So, as can be
S ( f n )∆f deduced from (16) the reference aerodynamic torque
σ{An } = σ{Bn } = (12) becomes a wind speed quadratic function or a
2
rotational speed quadratic function.
The equivalent wind speed time series can then
Another approach to generate a wind speed time
be applied to a specific turbine type, in which the
series according to a selected power spectral density
power coefficient and the rotor speed are controlled
model is based on the AR (Auto Regressive) or
variables.
ARMA (Auto Regressive Moving Average) models,
The aerodynamic torque can now be used for
[18].
simulation purposes as the input to the, more or less
The turbine rotor blades sweep through a
complex, turbine-(gearbox)-generator set mechanical
complex three dimensional wind field which varies
model.
both in time and over the rotor disc. How much the
3 Wind farm operation increased dynamic loading on the downstream
machines.
In a wind farm turbine wakes can interact, and
For wind farm studies, the all wind turbines must be
some turbines may be affected by the wakes of
considered. In large wind farms, at least some sub sets
several machines located upstream. This influence
must be considered in order to obtain realistic
depends on several parameters as farm geometry,
simulated results especially in what respects to power
wind speed and direction, and is depicted in Figure 5,
fluctuations and grid disturbances rejection. The
where dLo and dLa the longitudinal and lateral turbine
electrical power characteristics fundamentally depend
separation respectively.
on the average wind speed and turbulence. So, these
Wind farm codes usually rely on the results of
two wind parameters inside the all farm must be
single wake calculations, and use velocity deficits
known.
with linear or quadratic superposition to take into
Three important aspects must be considered in a
account the combined effect of different wakes.
wind farm simulation: the existence of some degree of
Experimental measurements and detailed simulations,
correlation between the wind speeds in different
([19] and [20]), have shown that when there are many
points in the farm; the possible operation of a wind
turbines in a line while the first turbine produces full
turbine under the wakes of others; and the coherence
power, there is a power decrease in the downstream
spectrum between two wind farm points.
turbines.
3.1 Correlation
In a wind farm with a large number of wind turbines
interconnected, smoothing effects are of great
importance for evaluating their output power
fluctuation, especially in variable speed turbines.
These smoothing effects are basically dominated by
the wind characteristics spatial correlation.
Since wind energy is a function of wind speed,
the establishment of a model for wind speed
correlation in a region is of great importance in the
management of wind energy resources for power
generation. The information obtained by this process
can be used to predict the wind power at some crucial
places of a wind farm in order to feed forward control
of the wind turbines for maximum performance or to
terminate its operation in case of an extreme gust.
Power quality measurements of wind turbines
connected to public grids are under the IEC 61400-21
standard, which specifies that the maximum 200 ms
average power as well as the maximum 1 min average
power must be measured. For the analysis of a wind
farm connection IEC 61400-21 assumes that the
turbulence of the wind at the different wind turbines is
uncorrelated. A more accurate measurement can be
obtained if it is considered the existence of a
correlation factor between the wind turbines in a farm.
3.2 Operation in wakes Figure 5: Turbine wakes in a wind farm with dLa=4D,
dLo=6D: a) predominant wind direction, b) one
A wind turbine extracts kinetic energy from the wind dimensional wake superposition and, c) two
and, in doing so, creates a wake. The wake has two dimensional wake superposition.
important characteristics: reduced mean wind speed
and increased turbulent energy. The reduced mean
wind speed will result in a reduced energy output of The wake velocity deficit decreases non-linearly
wind turbines operating in the wakes of others, [19]. with the distance; it is minimal beyond 10 rotor
The increased turbulence intensity results in an diameters downstream of a machine. Thus, the
commonly accepted practice in wind farm 0.32
D
arrangement is to space turbine rows between 6 and ∆I = 0.73a 0.83 I ∞− 0.0325 (20)
10 rotor diameters apart in the predominant wind x
direction and between 3 to 6 rotor diameters spacing
in the crosswind direction. where I ∞ is the ambient turbulence and a is a
A conservative spacing practice due to space parameter.
limitations or interconnection costs results in more
important wind power losses due to wake effects; its
influence should be considered.
Complete and detailed wake models are
necessary for structural and mechanical related
studies. For economic and grid connection assessment
more simple models can be used. Analytical
expressions characterizing the changes in the average
wind speed and in the turbulence are usually
considered.
The average wind speed change in the far wake
can be generally expressed in function of the distance,
(17), or in function of the transport time, (18).
The reduced wind speed in function of the
distance is given by:
n
∆U D Figure 6: Single wake average wind speed decrease in
= A (17)
Uh x the far wake for a range of parameters (rotor
frequency, fr, and thrust coefficient, Ct).
where Uh is the wind speed at hub height, x is the
downstream distance and D is the turbine rotor
diameter; A and n are constants. Constant A depends The actual turbulence intensity, I, is then given
on the turbine characteristics (thrust coefficient, Ct, by (21), [23]:
and the induced velocity factor, a), increasing with
both. These constants are in the range 1<A<3, and I = ∆I 2 + I ∞2 (21)
0.75<n<1.25.
The wind speed reduction related to the transport
time is given in [21] as: In dense arrangements the whole wind farm can
create its own inside ambient turbulence. Considering
the case of an infinitely large wind farm that perturbs
∆U t
= C2 ln 0 + Ct (18) the whole boundary layer, the estimated value of the
Uh t added turbulence intensity deep inside the wind farm
is obtained in [24] as:
where C2 is a constant and t0 is the transport time
where the near wake ends. Its expression is: 0.36
∆I = (22)
1 + 0.2 d r' d c' / Ct
1 H D
t0 = C1 ln (19)
f r z0 2 H where dr’ is the normalized (by the rotor diameter)
row distance and dc’ the normalized column distance.
where C1 is a constant, fr is the rotational frequency, The changes in turbulence intensity and average
z0 is the roughness height and H is the hub height. wind speed can be dealt with different expressions. Its
As a short example, Figure 6 shows the average applicability essentially depends on the considered
wind speed decrease in the far wake, obtained with distances, leading to near or far wakes.
(18) for two thrust coefficients and two rotational Distances between rows and between turbines in
speeds, considering C1=1.0, C2=0.4, H=55,z0=0.04, a row are other variables conditioning the increased
D=40. turbulence expressions validity, [22]. Also, complex
The added turbulence intensity created by a wind terrains introduce additional difficulties in wind
turbine can have different expressions, almost all modelling.
fitted from experimental measurements, [22]. One of Wind direction is other parameter influencing the
those expressions, (20), is presented in [23]: wind turbines operation in wakes, [19]. Ultimately,
the wind farm itself originates modifications in the in the separation distance. This other general three-
environment namely changing the roughness length dimensional exponential model is expressed in (25):
and thus creating a wind farm wake, [20].
− a ( fd / U ) 2 + (bd / Lc ) 2
γ2( f ) = e (25)
3.3 Turbulence inside a wind farm
where a=8.8 and b=0.12. Parameter Lc must be
The electrical power characteristics at the output of a
determined experimentally.
wind farm depend on all the individual wind turbines.
However, as demonstrated in [9] and [26], a large
Active power, flicker, feed forward control, and
wind farm needs a more accurate coherence model in
production forecasting accurate estimations all depend
order to get results with a high confidence level. In
on the wind farm behaviour as a whole and on each
modern wind farms the most important coherence
turbine operation. So, a wind speed time series is
factors are the longitudinal and the lateral ones.
needed at the input of each simulated turbine.
Generalization of (24) to obtain the coherence in a
A frozen turbulence hypothesis is a reasonable
plane can be done including the angle, α, between the
one for simulations where the simulated time series
wind direction and the line connecting the (i, j)
only pass the object once, like a wind turbine rotor.
reference points, (26):
For wind farm simulations, this condition is not
realistic, especially when the wind direction is along a
fd
line of wind turbines. In this case, simulations with a −σ (au / U cos(α) )2 + (av sin(α ) )2
frozen turbulence will generate wind speed time series γ2 ( f ) = e U (26)
which are identical for all wind turbines in the line,
with only a time delay associated with the travel time where au and av are the decay factors relative to the
between the two turbines. The assumption of a frozen longitudinal and lateral directions, respectively.
turbulence means that the coherence between the An example of the coherence factor evolution,
turbulence in two points on a line in the wind according to (26), is shown in Figure 7 for typical
direction will be one, which is not usually true. One wind speed characteristics (average speed, standard
important consequence of a frozen turbulence is that deviation and direction).
the summation of the power variations from the wind
turbines will be affected, [25].
The original Davenport empirical coherence
model used to calculate the cross power spectrum was
extended to the three directions. For the longitudinal
wind turbulence component it is described by an
exponential function, (23), with a decay parameter, au:
γ u2 ( f ) = e − au ( fd / U ) (23)