You are on page 1of 8

J. Wind Eng. Ind. Aerodyn.

109 (2012) 1–8

Contents lists available at SciVerse ScienceDirect

Journal of Wind Engineering


and Industrial Aerodynamics
journal homepage: www.elsevier.com/locate/jweia

Modeling of atmospheric wind speed sequence using a lognormal


continuous stochastic equation
R. Calif a,n, F.G. Schmitt b
a
EA 4098, LARGE laboratoire en Géosciences et énergétique, Université des Antilles et de la Guyane, 97170 P-á-P, Guadeloupe
b
CNRS, UMR 8187 LOG Laboratoire d’Océanologie et de Géosciences, Université de Lille 1, 28 avenue Foch, 62930 Wimereux, France

a r t i c l e i n f o abstract

Article history: In this paper, we have presented a spectral and a multifractal analysis performed on 412 time series of
Received 14 October 2011 wind speed data each of duration of 350 s and sampled at 20 Hz. The average spectrum for the wind
Received in revised form speed displays a scaling behavior, in the inertial range, over two decades, with b ¼ 1:68 close to the
10 May 2012
Kolmogorov value  5/3. A multifractal analysis has been motivated by the presence of scaling
Accepted 1 June 2012
invariance in data set. Then we have considered their scaling properties in the framework of fully
Available online 3 July 2012
developed turbulence and multifractal cascades. The results obtained for wind speed confirm that the
Keywords: exponent scaling function zV ðqÞ is nonlinear and concave. This exponent characterizes the scaling
Turbulence functions in the inertial range indicating that the wind speed is intermittent and multifractal. Moreover
Wind speed fluctuations
the theoretical quadratic relation for lognormal multifractals is well fitted. We investigate the
Intermittency
consequence for wind energy production: we generate stochastic simulations of a multifractal random
Multifractal analysis
Fractional Brownian motion walk, and using a power curve derived from experimental data, we generate the associated power time
Lognormal model series. We show that, due to the saturation of the power curve for large speed values, when the input
Multifractal random walk time series (turbulent wind speed) is multifractal, the output can be almost monofractal.
& 2012 Elsevier Ltd. All rights reserved.

1. Introduction turbulence, the most classical one being structure functions


analysis. In recent years several works have been dedicated to
Wind energy is a ubiquitous resource and a promising alter- the analysis of scaling laws and intermittency at small scales for
native to meet the increasing demand for energy. However, this speed in the laboratory (She and Levêque, 1994; Anselmet et al.,
energy production is very sensitive to fluctuations in wind speed, 1984) and in atmospheric boundary layer (Schmitt et al., 1993;
and these fluctuations are stronger for larger values of the wind Katul et al., 1995; Schmitt, 2007; Böttcher et al., 2007). These
speed, due to turbulence. Rapid variations of wind speed due to studies have shown that atmospheric turbulent speed at small
turbulence can lead (i) to electrical power variations in the order scales has multifractal scaling fluctuations and exhibits long
of the nominal power output, (ii) to mechanical stress on the gear range power correlations.
boxes and result in substantial voltage swings at the terminals In this paper, we characterize the scaling properties of high
(Fariley, 2003; Peinke et al., 2004). In this context, a precise frequency wind speed data sampled at 20 Hz in order to quantify
understanding and modeling of small scale turbulent wind field is and model the intermittency and statistical properties of their
important. fluctuations using a multifractal analysis technique considering
In the atmospheric boundary layer, the Reynolds Number the moment functions estimates. The paper is organized as
(ratio of inertial to viscous force) can be as large as 108 (Burton follows. In Section 2 the theoretical framework, traditional
et al., 2001). Large values of the Reynolds number lead to a huge spectral analysis and structure functions analysis are presented.
intermittency of wind speed fluctuations at all temporal or spatial In Section 3 the datasets are described. In Section 4 the results of
scales ranging from large-scale variations (years) to very small- the analysis are presented. In Section 5 the previous results are
scale variations (few minutes down to seconds). Small scale checked and confirmed using stochastic simulations.
intermittency remains a challenging problem for the turbulence
community research (Frisch, 1995). Several approaches can be
used to consider the scaling intermittency of small scale
2. Theoretical framework: turbulence and scaling

n
Corresponding author. We recall here in a few lines the basic scaling properties of
E-mail address: rcalif@univ-ag.fr (R. Calif). fully developed turbulence, first concerning Kolmogorov’s theory

0167-6105/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jweia.2012.06.002
2 R. Calif, F.G. Schmitt / J. Wind Eng. Ind. Aerodyn. 109 (2012) 1–8

(Kolmogorov, 1941), and then discussing scaling intermittency large fluctuations at all scales, with a correlated structure: large
descriptions and multifractal models. fluctuations are much more frequent than what would be
obtained for Gaussian processes (Frisch, 1995; Vulpiani and Livi,
2.1. Scaling properties following Kolmogorov (1941) 2004; Schertzer et al., 1997). This is typically studied considering
the probability density function (pdf) or the moments of the
The scaling and intermittent properties of wind speed have speed fluctuations, or more often, using the moment of order q of
been studied in the turbulence community for several decades these fluctuations, called ‘‘structure functions of order q’’
(see reviews in Frisch, 1995; Tennekes and Lumley, 1972; Monin
Sq ðtÞ ¼ /ðDV t Þq S C tzðqÞ ð4Þ
and Yaglom, 2007; Vulpiani and Livi, 2004; Schertzer et al., 1997).
We recall here some basic properties. Following the ideas of where we have written the fluctuations in time (t is a time scale),
Richardson and Kolmogorov for fully developed (and statistically since we deal with time series analysis in this paper. zðqÞ is the
homogeneous) turbulence, the Richardson–Kolmogorov energy scale invariant moment function, which is nonlinear and concave;
cascade develops on the inertial range between the large injection zð3Þ ¼ 1 is a fixed point; zð2Þ ¼ b1 relates the second order
scale (of the order of several tens or hundreds of meters) and the moment to the power spectrum scaling exponent. Knowledge of
dissipative scales (of the order of millimeters). In the inertial the full ðq, zðqÞÞ curve for integer and non-integer moments
range, wind speed fluctuations are scaling: they possess a power- provides a full characterization of speed fluctuations at all scales
law spectrum and all intensities. The parameter H ¼ zð1Þ is the Hurst exponent.
b Monofractal processes correspond to a linear function zðqÞ ¼ qH,
EðkÞ C k ð1Þ
where Brownian motion is described by H ¼ 1=2, fractional
with E the power spectral density, b ¼ 5=3 for K41 turbulence Brownian motion is described by 0 o H o1 (H a 1=2) (that can
(Kolmogorov, 1941; Obukhov, 1941) (k is the wave number). This be defined as a fractional integration of order b (0 o b o1) of a
relation is written here for wave numbers, but can be converted Gaussian noise, and H ¼ 32b) and homogeneous turbulence is
to frequencies using Taylor’s hypothesis (Tennekes and Lumley, described by H ¼ 1=3.
1972) to relate spatial fluctuations with temporal fluctuations. The values of the function zðqÞ are estimated from the slope of
Let us note that in the wind energy community, some fits of Sq ðtÞ versus t in a log–log diagram for all moments q. The function
speed power spectra are often used, based on von Karman’s zðqÞ defines the types of scaling behavior; in other words, this
2
formula of the form Eðf Þ ¼ a=ðbþ cf Þ5=6 (von Karman, 1948) or exponent function is useful to characterize the statistics of the
on Kaimal’s formula Eðf Þ ¼ A=ð1þ Bf Þ5=3 (Kaimal et al., 1972), random process. If zðqÞ is linear, the statistical behavior is
where a, b, c and A, B are constants. Von Karman wanted to monoscaling and if zðqÞ is nonlinear and concave, the behavior
perform an interpolation between a low-frequency f4 spectrum is defined as multiscaling, corresponding to a multifractal process.
5=3
and a high frequency Kolmogorov spectrum f and Kaimal The concavity of this function is a characteristic of the intermittency,
et al. proposed their formula on purely empirical grounds for the the more concave the curve is, the more intermittent the process
low frequency part. In both cases the main idea is to capture in a (Frisch, 1995; Vulpiani and Livi, 2004; Schertzer et al., 1997).
single expression the injections scales and the inertial range with
Kolmogorov spectrum. Indeed, these fits are based on Kolmogorov
theory (Simiu and Scanlan, 1978) which defines a general form of 3. Experimental dataset
the fluctuating wind velocity shown as follows (Zhangb et al.,
2008): In this study, we consider time series of speed V measured on
g the wind energy production site of Petit-Canal in Guadeloupe, an
fEðz,f Þ An
¼ ð2Þ island in the West Indies, located at 16115’N latitude and 60130’W
v2n ð1þ Bna Þb longitude. This 10 MW production site is positioned at approxi-
where f is the frequency, Eðz,f Þ is the spectrum, vn is the shear mately 60 m (197 ft) above sea level, at the top of a sea cliff. The
wind speed, n ¼ fz=vðzÞ is the Monin coordinate and vðzÞ is the wind speed is measured with an ultrasonic anemometer (model
mean wind speed. A, B, a, b and g are coefficients dependent on CSAT3) mounted on a tall mast erected 20 m (66 ft) from the cliff
the measured data and assumptions used in different models. edge, at 38 m (125 ft) from the ground. We analysed a database
Furthermore, the uncertainties and randomness are inherent in recorded at f s ¼ 20 Hz during 40 h in July 2005. We separated this
these power spectrum density functions (Solari and Piccardo, sequence into 412 series of length 350 s each, and perform
2001). In the present paper, we mainly focus on the inertial range ensemble average on the result, in order to characterize and
frequencies and therefore do not need to consider the fits model the properties of the atmospheric wind speed in the
proposed by these authors. inertial zone, i.e. for the small time scales. An example of atmo-
Kolmogorov’s scaling expression can also be written for spatial spheric wind speed sequence is displayed in Fig. 1. It shows a
fluctuations using the absolute increment DV ‘ ¼ 9Vðx þ ‘ÞVðxÞ9 strong intermittency, with fluctuations at all scales.
with V the wind speed, x defining the position, l a spatial distance. Let us characterize the turbulent flow using the speed time
Originally Kolmogorov proposed to consider the scaling series data and relations which are well-known in the inertial
properties of the second moment of these structure functions range of fully developed turbulence (Tennekes and Lumley, 1972).
(Kolmogorov, 1941): First, the dissipation E can be estimated by the following exact
g relation (Tennekes and Lumley, 1972; Frisch, 1995)
/ðDV ‘ Þ2 S C ‘ ð3Þ
/ðDV t Þ3 S ¼ 45EV 0 t ð5Þ
with the angle braces /  S indicating the ensemble average and
g ¼ 2=3 obtained from dimensional analysis. which is written here in the time domain, and considering that
the spatial distances ‘ writes ‘ ¼ V 0 t, where V0 ¼8.4 m/s is the
2.2. Intermittency and multifractal properties mean speed of the data collected during 40 h. The above relation
is used to obtain the dissipation: E ¼ 0:0027 m2 =s3 , which is quite
Intermittency in turbulence has been a subject of research for a large value. Then, the fluctuating speed or the standard devia-
almost 50 years now, following Kolmogorov’s (1962) seminal tion sv is estimated and can be used to determine the kinetic
work. Here we may define intermittency as the property of having energy k ¼ 3s2v =2. We find sv ¼ 1:95 m=s; this corresponds to a
R. Calif, F.G. Schmitt / J. Wind Eng. Ind. Aerodyn. 109 (2012) 1–8 3

14 105
Ev(f)
12 104 slope −1.68
slope −5/3
103
wind speed (m/s)

10

E(f) ((m/s)2/Hz)
102
8

101
6
100
4
10−1
2
10−2

0
10−3
0 2 4 6 8 10 12 14
10−3 10−2 10−1 100 101
time (hour)
f (Hz)
Fig. 1. An example of wind speed data sampled at 20 Hz and recorded continu-
ously during approximately 14 h, illustrating its intermittency. Fig. 2. The mean Fourier power spectrum EV ðf Þ (solid line) of the 412 samples of
wind speed data in log–log plot, showing a power-laws with a slope of  1.68
(line) compared to slope 5=3 (dashed line).
large turbulence intensity of sv =V 0 ¼ 22%. With these values we
can estimate the injection scale L ¼ s3v =E (Tennekes and Lumley,
1972), giving L¼2746 m. The Reynolds number can then be
estimated as Re ¼ V 0 L=n ¼ 2:57  1010 , where we have taken the demonstrates a scaling behavior for frequencies from about
kinematic viscosity n ¼ 0:896  106 m2 =s3 at 25 1C. This very f¼0.1 to f ¼10 Hz, with b ¼ 1:68 close to the Kolmogorov value
large value of the Reynolds number confirms the fully developed 5=3 (Kolmogorov, 1941; Obukhov, 1941): this is consistent with
turbulence framework. We can also estimate the Kolmogorov the values obtained for the inertial range in the previous studies
scale Z ¼ ðn3 =EÞ1=4 ¼ 0:13 mm; this corresponds to the smallest (Schmitt et al., 1993; Katul et al., 1995; Schmitt, 2007; Katul and
scale of turbulence. The inertial range is from the injection scale Chu, 1998; Lauren et al., 1999; Morales et al., 2012). The slight
to the Kolmogorov scale, or considering time fluctuations (and difference with the exact 5=3 value is usually caused by inter-
using Taylor’s hypothesis to relate time to space using the mean mittency effects (Frisch, 1995; Schertzer et al., 1997).
speed), from 326 s to 1.3 ms. pffiffiffiffiffiffiffiffiffiffiffiffiffi However, the power spectrum E(f) is second-order statistic
The Taylor microscale l is estimated as l ¼ 15n=Es ¼ 0:13 m. (proportional to the square of the amplitude of a given frequency
Finally, this allows us to estimate the Taylor-based Reynolds fluctuation) and its slope is not sufficient to fully specify a scaling
number Rl ¼ sv l=n ¼ 3  105 , which is much larger than the process. Multifractal analysis is a natural generalization to study
largest values of Rl obtained in the laboratory (Rl ¼ 104 ). the scaling behavior of a nonlinear phenomenon, using qth order
structure functions: this is discussed below.

4. Multifractal analysis of atmospheric wind speed


4.2. The structure functions scaling exponent zV ðqÞ for wind speed
We first confirm here the intermittent character of wind data
atmospheric fluctuations at small scales through multifractal
analysis. For that, the time period considered is two days sampled Here, we analyzed 412 time series of atmospheric wind speed
at 20 Hz for wind atmospheric measurements. V t ðtÞ each of duration 350 s and sampled at 20 Hz (ensemble
average is performed over 412 realizations). The structure functions
4.1. Spectral analysis analysis is realized for speed increments DV t ðtÞ ¼ 9Vðt þ tÞVðtÞ9 for
t belongs to the inertial range between 0.05 s and 1 s and for all
In order to estimate the power spectral densities of the wind moments between q¼0.15 and q¼5 with increment 0.25. Fig. 3
speed, the discrete Fourier transform of the times series consid- gives the scaling of the structure function for a sequence V t ðtÞ for
ered are computed. The expression of the power spectral density q¼1, 1.5, 2 and 2.5, in log–log plot. The straight lines in this figure
for a process x(t) is recalled here. A N point-long equal interval indicate that the scaling of the relationship (3) is well respected. We
time sample of process is used to construct the value at frequency then estimate zn ðqÞ for each time sequence n. The scale invariant
domain point f, Xf (Bracewell, 1999) function zV ðqÞ is the ensemble average obtained from all these
exponents zn ðqÞ
X
N 1
Xðf Þ ¼ xj e2pıEf =N ð6Þ
j¼0 1 X 412
zV ðqÞ ¼ z ðqÞ ð8Þ
412 n ¼ 1 n
with f ¼ 0; 1, . . . ,N1. The power spectral density is
2 Fig. 4 presents the empirical structure function scaling expo-
Eðf Þ ¼ 9Xðf Þ9 ð7Þ
nent zV ðqÞ compared to the linear model K41 (zðqÞ ¼ q=3) and the
Fig. 2 shows the average spectrum EV(f) for the wind speed, lognormal model. The lognormal multifractal model is a classical
obtained as an ensemble average estimated over 412 sections model first proposed by Kolmogorov (1962) and still advocated
compared to 5=3 slope. The wind speed spectrum, EV(f) in recent years (Arneodo et al., 1996); it corresponds to the
4 R. Calif, F.G. Schmitt / J. Wind Eng. Ind. Aerodyn. 109 (2012) 1–8

Following these results, the goal is to reproduce the scaling


q=1
q=1.5 law for the moments and the power-law correlations of atmo-
q=2 spheric wind speed. Thus, a lognormal continuous stochastic
100 equation is considered in Section 5.
q=3
Fit q=1
Fit q=1.5
Fit q=2 5. Stochastic modeling of wind speed for wind power
Fit q=3
<|Δv|q>

10−1 fluctuations from one turbine

The fluctuating character of the output power from a Wind


Turbine Generator (WTG) is a consequence of the intermittency of
10−2 the atmospheric wind speed. The passage from the wind field to
the output power is usually modeled using a power curve. In
order to test the smoothing effect of a wind turbine, we model the
wind turbine behavior using a power curve. For that, we perform
in this section stochastic simulations of wind speed time series,
10−3
and analyze the scaling properties of the output power time series
0.05 0.5 5
using a virtual wind turbine using a specific power curve. In this
τ (s)
section, we model the turbulent wind speed using two scaling
Fig. 3. The scaling of the structure functions for wind speed SV(q), for q ¼1, 1.5, stochastic models: a fractional Brownian motion (fBm) and a
2 and 2.5. multifractal non-stationary sequence.

5.1. Stochastic equation for monofractal processes: fractional


2 Brownian motion
ζv
1.8 Introduced by Kolmogorov (1940), Yaglom (1957) and fully
q/3
Log−normal described in other studies by Mandelbrot and Van Ness (1968),
1.6
fBm is a fractional integration of a Gaussian white noise process. A
1.4 fBm time series X(t) is a Gaussian time series whose autocorrela-
tion function is given by
1.2
s2 2H
ðt 2H þ s2H 9ts9
ζv(q)

1 Rðt,sÞ ¼ EðXðsÞXðtÞÞ ¼ Þ ð10Þ


2
0.8 where s is the standard deviation of X(t) and the Hurst parameter
H A ½0; 1. The fBm is a monofractal process which has stationary
0.6 increments and has scaling structure function following zðqÞ ¼ qH.
0.4
5.2. Multifractal random walk simulation
0.2
To generate a non-stationary multifractal process mimicking
0
the turbulent speed, we generalize multifractal random walk
0 1 2 3 4 5
q ideas (Bacry et al., 2001; Muzy and Bacry, 2002), and generate a
non-stationary multifractal time series u(t) as
Fig. 4. Empirical values of the structure function scaling exponent zv ðqÞ for wind Z t
speed (&) compared to the linear model K41 (dashed line) and the lognormal uðtÞ ¼ E1=2 ðxÞ dBH ðxÞ ð11Þ
model (solid line). 0

where BH(x) is a fractional Brownian motion. Taking lognormal


statistics for E, an intermittent and long-range correlated time
non-linear multifractal curve
series (Schmitt, 2003), the scaling exponent zðqÞ given by
q m q
/9Dut ðtÞ9 S  tzðqÞ can be written as
zðqÞ ¼ þ ð3qq2 Þ ð9Þ
3 18
m
zðqÞ ¼ qH ðq2 qÞ ð12Þ
where the parameter m is called ‘‘intermittency parameter’’; the 2
value m ¼ 0:25 70:05 has been proposed (Arneodo et al., 1996). where m is the intermittency parameter (0 r m r1) characterizing
The function zV ðqÞ obtained from our database is nonlinear and the lognormal multifractal cascade. This was done in Huang et al.
concave similar to the results obtained in previous studies (2008) for H ¼ 1=2; we use H o 1=2 here. This multifractal
(Schmitt et al., 1993; Schertzer et al., 1997) has the following stochastic process is continuous in scale ratio and in time. It has
properties: a simple generating equation and can be used to sequentially
generate time series of any length.
 0 o qo 3, zV ðqÞ 4 q=3 for small fluctuations,
 q¼3, zV ð3Þ ¼ 1 in agreement with the Kolmogorov 4=5 law, 5.3. Stochastic simulations of atmospheric wind speed data and
 q4 3, zV ðqÞ o q=3 for large fluctuations. wind power through the power curve

The lognormal model provides a reasonable fit for the empirical Here, we simulate a sequence of length 4096 data points for a
exponent function zV ðqÞ. Other models have been proposed such fBm process with H ¼ 1=3 using the circulant matrix method
as log-Poisson (She and Levêque, 1994) and log-stable (Schertzer algorithm (Wood and Chan, 1994). For lognormal multifractal
et al., 1997), but we do not test them here. process, we provide a stochastic simulation for a sequence of
R. Calif, F.G. Schmitt / J. Wind Eng. Ind. Aerodyn. 109 (2012) 1–8 5

length 4096 with m ¼ 0:25. The sampling frequency of the simu- random walk. Gaussian statistics are compatible with fBm, con-
lated data is 20 Hz. Then, the spectra and the exponent structure trary to the measured signal and the multifractal random walk.
functions obtained in each case are compared with the empirical In Fig. 7, we have plotted the spectra of measured and
spectrum E(f) and curve zv ðqÞ of a measured wind speed signal simulated speed signals compared to a straight line of slope
presented in Fig. 1. The simulated signals have been rescaled to b ¼ 5=3: the scaling behavior is verified. We checked also the
have the same mean value, 6 m/s (this value is the most probable structure functions scaling exponents of vfBm and vmrw. Fig. 8
occurring on the site Calif et al., 2005), and the same standard shows the scaling exponents of vfBm, vmrw and vexp compared to
deviation 0.3 m/s as the measured signal vexp. Fig. 5 illustrates the the linear model K41 (zðqÞ ¼ q=3Þ and lognormal model
measured signal vexp, a multifractal random walk vmrw generated (zðqÞ ¼ q=3 þðm=18Þð3qq2 Þ). For the signal generated with frac-
by the lognormal continuous stochastic equation and a fractional tional Brownian motion, the curve zfBm ðqÞ is linear and fitted by q/
Brownian motion vfBm generated by the circulant matrix method 3. The curve of zmrw ðqÞ of the generated signal with lognormal
algorithm. Fig. 6 shows the normalized densities of measured multifractal random walk is superimposed to empirical curve
signal Dvexp , fBm process DvfBm and multifractal random walk zv ðqÞ. zmrw ðqÞ is nonlinear and concave: the field that is generated
Dvmrw , compared to the Gaussian pdf G  Nð0; 1Þ for the temporal with this type of equation is intermittent, scale invariant, and
increments Dv ¼ vðt þ tÞvðtÞ with t ¼ 0:5 s. We observe that characterized by multiple-scaling properties. Such properties are
large fluctuations have larger probabilities in the multifractal fully compatible with experimental results.

7
vexp (m/s)

6.5
6
5.5
0 20 40 60 80 100 120 140 160 180 200

7
vmrw (m/s)

6.5
6
5.5
0 20 40 60 80 100 120 140 160 180 200

7
vfBm (m/s)

6.5
6
5.5
0 20 40 60 80 100 120 140 160 180 200
time (seconds)

Fig. 5. (a) An example of measured signal vexp, (b) a multifractal random walk vmrw generated by the lognormal continuous stochastic equation and (c) a fractional
Brownian motion vfBm generated by the circulant matrix method algorithm.

100 100 100

10−1 10−1 10−1


density

10−2 10−2 10−2

10−3 10−3 10−3


−5 0 5 −6 −4 −2 0 2 4 6 −5 0 5
Δvexp/σ Δvmrw/σ ΔvfBm/σ

Fig. 6. The increment normalized densities of Dvexp ðtÞ (left), Dvmrw ðtÞ (center) and DvfBm ðtÞ (right), compared to the Gaussian pdf G  Nð0; 1Þ.
6 R. Calif, F.G. Schmitt / J. Wind Eng. Ind. Aerodyn. 109 (2012) 1–8

250
Emrw
103 Eexp
Efbm
102 slope −5/3 200

101
E(f) ((m/s)2/Hz)

150

P (kW)
100

10−1 100

10−2
50
10−3

10−4 0
10−3 10−2 10−1 100 101 0 5 10 15 20 25 30 35
f (Hz) v (m/s)

Fig. 7. The Fourier power spectrum E(f) of wind speed (Eexp), fBm (Efbm) and Fig. 9. Power curve providing the power curve of a 220 kW wind turbine
lognormal multifractal random walk (Emrw) compared to slope 5=3. generator versus the input wind speed.

1.8 In order to analyze the scaling law of the moments for the
q/3 output power from one wind turbine, we generate the output
1.6 ζvfBm power with the wind speed modeling results for 4o vðtÞ r15 m=s
Log−normal (increasing zone of the power curve) and 15 o vðtÞ r26 m=s (stall
1.4 region of the WTG).
ζv
1.2 ζvmrw
5.3.1. Modeling for wind power in the increasing zone of the power
1 curve
ζ(q)

We first consider here the increasing zone of the power curve,


0.8
for which 4 o vðtÞ r15 m=s. We study three simulated power
0.6 series: (i) P simul is generated with the experimental wind speed
data vexp; (ii) Pfbm is simulated with the wind speed modeling
0.4 using a fractional Brownian motion data vfbm; and (iii) Pmrw is
generated with the wind speed modeling using a lognormal
0.2 multifractal random walk vmrw. We simulated 20 segments of
4096 points sampled at 1 Hz for all three simulated power series.
0 Fig. 10a illustrates the Fourier power spectra EPsimul, EPfBm, EPmrw
0 1 2 3 4 5
for all three simulated power series. The power spectra of the
q
simulated power series follow a power-law spectrum for
Fig. 8. The function zv ðqÞ of wind speed zvexp (&), fBm zvfBm ðXÞ and lognormal 0:001 rf r 0:5 with a slope close to 5=3.
multifractal random walk zvmrw (n) compared to lognormal model zvlognormal (solid To determine the nature of scaling for Psimul, PfBm and Pmrw, we
line) and linear model K41 (dashed line). estimated the structure functions of their time increments and
the exponent structure functions zPsimul , zPfbm , zPmrw plotted in
The wind speed simulation using two stochastic model (frac-
Fig. 10b. For all three simulated power series, the scaling behavior
tional Brownian motion and multifractal random walk) are
of the output power is similar to that of the wind speed modeling
translated into wind power using the mean power curve of a
results obtained in Section 5.3. Indeed, Pfbmhas the same mono-
220 kW wind turbine currently installed and operating in the
scaling behavior as vfBm, whereas Psimul and Pmrw have the same
wind farm described in Section 3: this type of turbine possesses
multiscaling behavior as respectively vexp and vmrw. These results
two blades of length 13 m, with stall-regulated and fixed speed.
show that the spectral and statistical properties of the output
The goal is to see how wind speed modeling affects wind power
wind power are similar to those of the wind speed. In other
output, a point that is particularly important for the design and
words, although the power curve is a cubic function, the inter-
operation of a wind turbine. The turbines have a cut-in speed of
mittency of wind speed is unchanged by the power curve in this
4 m/s, a cut-out speed of 26 m/s and produce their peak power
region contrary to modeling of the output wind power using
output at 15 m/s. The power curve of the wind turbine generator
Langevin equation presented in Gottschall and Peinke (2007) in
is shown in Fig. 9. To simulate wind power P(t), we fit the
which the intermittency of the speed increment distribution
following power curve to experimental data collected and pro-
seems to be amplified.
cessed according to the IEC standard 61400-12 (Burton et al.,
2001):
5.3.2. Modeling for wind power in the stall region of the power curve
8 In order to analyze the intermittency of the output power in
2 3
< 13063vðtÞ þ 9:1v ðtÞ0:3v ðtÞ
> for 4 ovðtÞ o 15 m=s
the stall region of the power curve, we simulated 20 segments
PðtÞ ¼ 140 þ 11vðtÞ0:36v 2
ðtÞ for 15 r vðtÞ o 26 m=s of wind speed having a mean value around 16 m/s, using a
>
:0 for vðtÞ Z 26 m=s lognormal multifractal random walk. The spectral and
R. Calif, F.G. Schmitt / J. Wind Eng. Ind. Aerodyn. 109 (2012) 1–8 7

105 1.8
EPfBm ζPsimul
1.6 Log−normal
104 EPmrw
q/3
slope −5/3 1.4 ζPfBm
103 EPsimul
ζPmrw
1.2
E(f) (kW2/Hz)

102 1

ζ(q)
101 0.8

0.6
100
0.4
10−1
0.2

10−2 0
10−4 10−3 10−2 10−1 100 0 1 2 3 4 5
f (Hz) q

105 1.8
EPS ζPS
slope −1.64 1.6 ζPL
EPL
slope −1.63 q/3
1.4

1.2
E(f) (kW2/Hz)

1
ζp(q)

100
0.8

0.6

0.4

0.2

10−5 0
10−4 10−3 10−2 10−1 100 0 1 2 3 4 5
f (Hz) q

Fig. 10. (a) The Fourier power spectrum E(f) for P simul (EPsimul), Pfbm (EPfbm) and Pmrw (EPmrw), compared to slope 5=3; (b) the function zp ðqÞ for simulated wind power from
speed data zPsimul (&), fBm zPfbm ðXÞ and lognormal multifractal random walk zPmrw (solid line) compared to fit q/3 (dotted-line); (c) the Fourier power spectrum of
simulated wind power from speed generated using lognormal multifractal random walk, for the linear zone EPL and EPS for the stall region of the power curve (saturated
zone of power curve); (d) the function zp ðqÞ for 10 simulated samples of wind power from speed synthesized using lognormal multifractal random walk, for the linear zone
zPL and stall zone zPS .

statistical properties of Ps are compared with those obtained for 6. Conclusions


Pmrw in the increasing zone of the power curve. Fig. 10c
illustrates the Fourier spectrum Eps(f) for Ps and EpL(f) for Pmrw. We considered here wind speed time series sampled at 20 Hz.
The two spectra have a 5=3 power spectral slope for Their scaling statistics have been studied in the framework of
0:001r f r 0:5. Moreover, as shown in Fig. 10c, the variance of fully developed turbulence and Kolmogorov’s theory. Their Four-
Pmrw is larger than that of Ps. We estimate the structure function ier power spectra have been estimated and we considered their
for the time increment of Ps for determine the exponent scaling properties in the physical space. We showed that the wind
function zps ðqÞ. In Fig. 10d, we have plotted zps ðqÞ (&) and speed is intermittent and multifractal as previous works. In order
zpmrw ðqÞ (1) compared to linear fit q/3 (solid line). zpmrw ðqÞ is to check this results, we generated turbulent wind speed time
nonlinear and concave whereas zps ðqÞ tends to be linear. Indeed, series using two stochastic models, fractional Brownian motion
the figure shows a slight difference between zps ðqÞ and the fit q/ (fBm) and multifractal random walk. The analyses have shown
3. In the stall region, the large fluctuations of the wind speed that the multifractal random walk is intermittent, and possesses
are reduced for the wind power fluctuations. scaling properties very similar to measured data, unlike fBm.
In conclusion the behavior of the output fluctuations seems Using these stochastic simulations and a measured signal, we
multifractal for the increasing range and monofractal for the provide a simulation for wind power data: the wind speed
stalled range. However the lack of real data for a single turbine simulation results and the wind speed measured are translated
does not enable us to conclude in a final way. into wind power using the power curve of one of the 220 kW
8 R. Calif, F.G. Schmitt / J. Wind Eng. Ind. Aerodyn. 109 (2012) 1–8

wind turbines we collected data from. The simulated power Calif, R., Blonbou, R., Deshaies, B., January 2005. Wind speed analysis for time
presents two scaling behaviors: (i) in the increasing region of scales smaller than 1 hour: application to wind energy forecasting. In: 24th
AIAA-ASME Wind Energy Symposium, Reno, 1332.
the power curve, P is a multifractal random walk and exhibits a Fariley, P., 2003. Steady as she blows. IEEE Spectrum 12, 35–39.
long-range power correlations according to Gottschall and Peinke Frisch, U., 1995. Turbulence the Legacy of A.N. Kolmogorov. Cambridge University
(2007), and (ii) in the stall region of the power curve, P is close to Press, Cambridge.
Gottschall, J., Peinke, J., 2007. Stochastic modelling of a wind turbine’s power
a monofractal random walk. In other words, the wind power is
output with special respect to turbulent dynamics. Journal of Physics:
intermittent and multifractal in the increasing zone, whereas the Conference Series 75, 012045.
wind power is less intermittent and close to monofractal process Huang, Y., Schmitt, F.G., Lu, Z., Liu, Y., 2008. An amplitude–frequency study of
in the stall zone, meaning that the large fluctuations of the wind turbulent scaling intermittency using Hilbert spectral analysis. Europhysics
Letters 84, 40010.
speed are smoothed in the stall region. Although no data for an Kaimal, J.C., Wyngaard, J.C., Izumi, Y., Cote, O.R., 1972. Spectral characteristics of
individual turbine are available, we attempt to give different surface-layer turbulence. Quarterly Journal of the Royal Meteorological Society
possible explanations concerning the smoothing effect. Besides, 98, 563–589.
Katul, G., Chu, C.R., 1998. A Theoretical and experimental investigation of energy-
we could only obtain the data corresponding to the aggregate
containing scales in the dynamic sublayer of boundary-layer flows. Boundary-
power of an entire wind farm. The dataset has been analyzed in Layer Meteorology 86, 279–312.
the same way (spectral analysis and multifractal analysis). The Katul, G.G., Chu, C.R., Parlange, M.B., Albertson, J.D., Ortenburger, T.A., 1995. Low-
analysis results published in Calif and Schmitt (2012) have shown wavenumber spectral characteristics of speed and temperature in the
atmospheric surface layer. Journal of Geophysical Research 100 (D7),
that the aggregate output power possesses intermittent and 14,243–14,255.
multifractal properties. Kolmogorov, A.N., 1940. The Wiener spiral and some other interesting curves in
This work thus predicts that output power time series possess 3218 Hilbert space. Doklady Akademii Nauk SSSR 26, 115–118.
Kolmogorov, A.N., 1941. The local structure of turbulence in incompressible fluid
intermittent and scaling statistics in relation with the multifractal
for very large Reynolds numbers. Doklady Akademii Nauk SSSR 30, 299–303.
properties of the input wind speed series. Furthermore, we Kolmogorov, A.N., 1962. A refinement of previous hypotheses concerning the local
numerically found that the saturation behavior of the nonlinear structure of turbulence in a viscous incompressible fluid at high Reynolds
transfer curve has an influence on the output power series: for an number. Journal of Fluid Mechanics 13, 82–85.
Lauren, M.K., Menabe, M., Seed, A.W., Austin, G.L., 1999. Characterisation and
absolute wind speed not too large, the power series will also be simulation of the multiscaling properties of the energy-containing scales of
multifractal whereas the effect of the saturation has the conse- horizontal surface layer winds. Boundary-Layer Meteorology 90, 21–46.
quence of producing non-intermittent in the power series. Mandelbrot, B.B., Van Ness, J.W., 1968. Fractional Brownian motions, fractional
3253 noises and applications. SIAM Review 10, 422.
Future works involve simultaneous analysis of turbulence time Monin, A.S., Yaglom, A.M., 2007. Statistical Fluid Mechanics, vol. II. MIT Press,
series and instantaneous power data, in order to consider the reprint Dover, p. 92.
multiscale correlation between both series. With the increasing Morales, A., Wächter, S., Peinke, J., 2012. Characterization of wind turbulence by
importance of wind energy, the understanding of the non-sta- higher-order statistics. Wind Energy 15 (3), 391–406.
Muzy, J., Bacry, E., 2002. Multifractal stationary random measures and multifractal
tionary and intermittent nature of power series is an important random walks with log infinitely divisible scaling laws. Physical Review E 66
issue. The present study shows that this can be related to (5), 056121.
turbulent wind speed properties in the framework of locally Obukhov, A.M., 1941. Spectral energy distribution in a turbulent flow. Doklady
Akademii Nauk SSSR 32, 22–24.
homogeneous Kolmogorov turbulence. Peinke, J., Barth, S., Bottcher, F., Heinemann, D., Lange, B., 2004. Turbulence, a
challenging problem for wind energy. Physica A 338, 187–193.
Schertzer, D., Lovejoy, S., Schmitt, F., Chigirinskaya, Y., Marsan, D., 1997. Multi-
Acknowledgment fractal cascade dynamics and turbulent intermittency. Fractals 5 (3), 427–471.
Schmitt, F.G., 2003. A causal multifractal stochastic equation and its statistical
properties. European Physics Journal B 34, 85–98.
This study is financially supported by the Regional Council of Schmitt, F.G., 2007. Gusts in intermittent wind turbulence and the dynamics of
Guadeloupe and the European funding (No 1/1.4/-31614). their return times. In: Peinke, J., Schaumann, P., Barth, S. (Eds.), Wind Energy,
Proceedings of the Euromech Colloquium. Springer, pp. 73–79.
Schmitt, F., Schertzer, D., Lovejoy, S., Brunet, Y., 1993. Estimation of universal
References multifractal indices for atmospheric turbulent speed fields. Fractals 1,
568–575.
Anselmet, F., Gagne, Y., Hopfinger, E.J., Antonia, R.A., 1984. High order velocity She, Z.S., Levêque, E., 1994. Universal scaling laws in fully developed turbulence.
structure functions in turbulent shear flows. Journal of Fluid Mechanics 140, Physical Review Letters 72, 336–339.
63–89. Simiu, E., Scanlan, R.H., 1978. Wind Effects on Structures: An Introduction to Wind
Arneodo, A., Baudet, C., et al., 1996. Structure functions in turbulence, in various Engineering. Wiley, New York.
flow configurations, at Reynolds number between 30 and 5000, using Solari, G., Piccardo, G., 2001. Probabilistic 3-D turbulence modeling for gust
extended self-similarity. Europhysics Letters 34, 411–416. buffeting of structures. Probabilistic Engineering Mechanics 16 (1), 73–86.
Bacry, E., Delour, J., Muzy, J.F., 2001. Multifractal random walks. Physical Review E Tennekes, H., Lumley, J.L., 1972. A Fit Course in Turbulence. The MIT Press.
64, 026103. von Karman, T., 1948. Progress in the statistical theory of turbulence. Proceedings
Böttcher, F., Barth, St., Peinke, J., 2007. Small and large fluctuations in atmospheric of the National Academy of Sciences of the United States of America 34,
wind speeds. Stochastic Environmental Research and Risk Assessment 21, 530–539.
299–308. Vulpiani, A., Livi, R. (Eds.), 2004. The Kolmogorov Legacy in Physics. Springer,
Bracewell, R., 1999. The Fourier Transform and Its Applications, third ed. McGraw- Berlin, Heidelberg, New York.
Hill Science. Wood, A., Chan, G., 1994. Simulation of stationary Gaussian processes. Journal of
Burton, T., Sharpe, D., Jenkins, N., Bossanyi, E., 2001. Wind Energy Handbook. John Computational and Graphical Statistics 3, 409–432.
Wiley & Sons, Chichester. (p. 12). Yaglom, A.M., 1957. Some classes of random fields in n-dimensional space, related
Calif, R., Schmitt, F.G., 2012. Analyse des séries temporelles de production 3441 to stationary random processes. Theory of Probability and Its Applica-
éolienne: loi de Taylor et propriétés multifractales, in Rencontre du non tions 2, 273. (3442).
linéaire 2012 éd. In: Josserand, C., Lefranc, M., Letellier, C. (Eds.), Institut Henri Zhangb, L., Lia, J., Pengb, Y., 2008. Dynamic response and reliability analysis of tall
Poincaré, Paris, Non-linéaire Publications, 2012. ISBN 978-2-9538596-1-4, buildings subject to wind loading. Journal of Wind Engineering and Industrial
pp. 61–66. Aerodynamics 96, 25–40.

You might also like