Renewable Energy 103 (2017) 15e29

Contents lists available at ScienceDirect

Renewable Energy
journal homepage: www.elsevier.com/locate/renene

Two and three-parameter Weibull distribution in available wind

power analysis
Piotr Wais
Institute of Power Engineering, Cracow University of Technology, al. Jana Pawla II 37, 31-864 Cracow, Poland

a r t i c l e i n f o a b s t r a c t

Article history: Based on the current EU environmental policy, wind power industry has been growing fast in recent
Received 20 November 2015 years. Knowledge of wind characteristics helps to define site requirements, choose a proper turbine
Received in revised form design and estimate profits from the wind energy production.
17 September 2016
The paper describes and compares the techniques of available wind energy evaluation using Weibull
Accepted 17 October 2016
distributions: two-parameter Weibull probability distribution, and three-parameter Weibull distribution.
Available online 25 October 2016
The two-parameter Weibull distribution is recognized as an appropriate model and the most widely used
in the wind industry sector. In some cases, in which the probability of null wind is significant, the Weibull
Keywords:
Wind energy calculation
distribution cannot reveal good conformity for the low wind speed. In theory, it seems that the three-
Two-parameter Weibull distribution parameter Weibull distribution, which takes into account the frequency of null winds, may better
Three-parameter Weibull distribution represent wind ranges for the low wind speed.
Coefficient of determination In the paper, the available wind energy is calculated for three different turbine locations. The results
Fit standard error show that for the higher probability of the null wind the three-parameter Weibull distribution gives
Null wind speed better results comparing to the two-parameter Weibull distribution and can be proposed as an alter-
native to wind energy estimation technique. Additional analyses confirm the observation.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction the wind turbine works. The energy captured from the wind for the
turbines that operate at the area with the poorer wind resources
Energy consumption continues to increase in our lives. To meet can be maximised. But still, the cost of the construction of wind
the demand and supply the necessary amount of energy, new po- farms and the payback period should be firstly determined [1]. The
wer plants are built or modernised. Based on the current EU energy output increment per additional meter of the hub height
environmental policies, which support and promote initiatives that falls down while the investment cost grows progressively. In the
generate energy from renewable sources, there is a large increase in past decade, the global wind industry has been still growing, and its
the construction of the wind power plants. Construction of new participation in the energy production increases every year. Despite
wind power is also connected with the requirements of energy new technical solutions and more efficient energy conversion, the
security and lower dependences on imported fossil fuels. For these turbine system cost and economic analyses are one of the most
reasons, wind energy plays an important role in the energy system important factors to evaluate the advantage or disadvantages be-
and is an alternative to conventional energy plants. tween wind and conventional power systems.
There are many different wind farms that vary in wind turbine The turbine energy production depends mostly on the wind
power, wind turbine type, the height of the tower, etc. To construct capacity and wind energy can be an attractive option, especially for
the wind power plant, it is necessary to invest relatively large sums sites with good stability in wind speed values and wind direction.
of money that depends on the turbine type and the tower con- Estimating the wind characteristics and having the wind turbine
struction. More energy can be generated at higher altitudes due to power curve, it is possible to calculate the energy production,
the fact that the average wind speed increases with the height of assess necessary investments and profits. The best way to assess
the hub. Ongoing innovations in turbine design help to improve the the wind energy is to carry out meteorological measurements for
turbine efficiency and widening the range of wind speed in which each proposed turbine location. Unfortunately, that activity re-
quires long time, and the measurement cost is remarkable. On the
other hand, the wind energy can be determined using statistical
E-mail address: [email protected]. analysis. In that case, the wind speed variation is characterized by a

http://dx.doi.org/10.1016/j.renene.2016.10.041
0960-1481/© 2016 Elsevier Ltd. All rights reserved.
16 P. Wais / Renewable Energy 103 (2017) 15e29

probability distribution function and to calculate wind power, it is percentages of null wind speeds. The typical two-parameter Wei-
necessary to find distribution parameters. Statistical analysis can bull wind speed distribution and the two component mixture
also be helpful to estimate and study the available wind energy in Weibull distribution with five parameters (two shape parameters,
regions, for which wind data are not available. two scale parameters, and one proportionality parameter) are
Reviewing the literature [2e12], it can be noticed that the two- selected distributions for examination. The general conclusion is
parameter Weibull probability function is a useful tool for wind that the mixture Weibull distribution is more suitable for the
energy estimation. However, that distribution may not be appro- description of such wind conditions. Carta and Ramirez [7] show
priate for properly fitting the wind data of low speed. In this paper, that a two-component mixture Weibull distribution is more useful
an investigation is made on the suitability of the Weibull proba- because it can also represent heterogeneous wind regimes, in
bility distribution functions to fit wind speed data and choose a which there is evidence of bimodality. An analysis is made of the
more accurate method for evaluation of wind speed characteristics. most frequently used methods for the estimation of the five pa-
Firstly, the analysis focuses on the typical two-parameter Weibull rameters of the Weibull distribution. The least-square method is
wind speed distribution. Then, the three-parameter Weibull proposed as a robust and computationally efficient alternative to
probability function is applied to match the statistical wind data the techniques currently in use. Arslan et al. [16] study the nu-
received from three meteorological stations in Poland that may merical method for determining Weibull parameters for wind en-
better correspond to the lower speed wind data and give more ergy potential. The L-moment method is presented and compared
appropriate results. The two and three-parameter Weibull distri- to the moment method and maximum likelihood method. To
butions are compared directly for wind energy calculations. The compare the estimation methods, the mean square error criteria
main purpose of the work is to check whether or not the three- are utilized. It is noted that in comparison to the other methods, the
parameter Weibull distribution can take advantages comparing to L-moment method yields better results with small sample sizes.
the typical Weibull distribution for high percentages of null wind Yaniktepe et al. [17] investigate the wind energy potential and the
speeds. wind characteristics using the two-parameter Weibull and Rayleigh
distribution methods. A graphical method is applied to determine
2. Overview of wind speed distribution function Weibull parameters. Analyses of wind characteristics are conducted
at different temporal scales, such as yearly and monthly. They
A proper statistical analysis of wind data is an important phase demonstrate that these probability distributions are frequently
in a wind resource evaluation. The selection of the wind speed used for this type of the energy study. Seguro and Lambert [4] es-
distribution function influences the calculation of the available timate the parameters of the Weibull wind speed distribution by
wind energy or the wind turbine performance at chosen locations, the maximum likelihood method, the proposed modified
and gives the significant impact on the investment profitability. maximum likelihood method, and the commonly used graphical
Wind speed frequency distribution is represented by various method. They show that the two-parameter Weibull distribution
probability density functions including the gamma function, the function is commonly used to fit the wind speed frequency distri-
beta function, the lognormal function, Rayleigh and Weibull dis- bution and provides a convenient representation of the wind speed
tributions. Analysing the literature and international standards data for wind energy calculation purposes. Carta et al. [11] review
[13e15], it can be seen that two-parameter Weibull distribution is the wind speed probability distributions used in wind energy
most commonly and widely used mathematical model to estimate analysis. They demonstrate that the most popular methods for
available wind energy in practical engineering applications. Also, it distribution parameter estimation are the moment method, the
is used for the energy calculations in commercial programs. maximum likelihood method and the least squares method. They
Therefore, a number of studies are carried out to assess the wind notice that the two-parameter Weibull distribution function is the
energy by means of the Weibull distribution. When fitting the data, most widely used and accepted distribution in the specialised
various methods to calculate the unknown distribution parameters literature on wind energy and presents a series of advantages.
of the Weibull distribution are proposed in different papers. Lun Ramirez and Carta [18] estimate the parameters of the two-
and Lam [3] calculate the two parameters of a Weibull density parameter Weibull wind speed probability density function using
distribution function for three different locations. Based on a long- different methods. They analyse the influence of the data sampling
term data source it was found that the numerical values of the interval in the estimation of the Weibull function parameters. They
shape and scale parameters for these weather stations varied over a express that the most frequently used are the method of moments,
wide range. Celik [5] derives probability density distributions from the maximum likelihood method and the least square method.
time-series data and calculates distributional parameters. Two Chang [19] evaluates the wind energy potential using different
functions for fitting a measured wind speed probability distribution probability density functions. He analyses the conventional Weibull
are the Weibull and Rayleigh functions. The month to month function with the bimodal Weibull function, truncated normal
comparison shows that the Weibull model better fits the measured Weibull function, the mixture GammaeWeibull function, the
probability density distributions comparing to the Rayleigh model. mixture truncated normal function and the probability density
Also, the Weibull model provides better power density evaluations. function derived with maximum entropy principle. The results
Wais [1] underlines that the correct estimation of Weibull param- show that all the mixture functions and the maximum entropy
eters is essential for the wind energy assessment. The information function describe wind characterizations better than the conven-
about the average wind speed in the region is not enough to tional Weibull function. Qin et al. [20] improve mixture Weibull
determine the wind speed distribution and to calculate energy distribution models by adding one or two location parameters to
production because it is possible to have different parameters of the existing two-component mixture two-parameter Weibull dis-
Weibull distribution for the same average wind speed. He studies tribution. The parameters of new models are estimated by the
four situations and demonstrates that for the average wind speed maximum likelihood method. The experimental result show that
3.75 m/s, and the rotor diameter 90 m, the calculated energy pro- the mixture Weibull models are more flexible than existing models
duction can vary even by 80%. Akdag et al. [9] analyse the statistical for the wind speed data analysis. Pishgar-Komeh et al. [21] analyse
characteristics of wind speed data in order to obtain a more accu- the wind speed data in one of the region in Iran. Results show that
rate method for estimation of wind speed characteristics. The two parameter Weibull distribution and Rayleigh distribution
article focusses on wind regimes that present nearly zero functions can fit the values of wind seed with almost the same
 P. Wais / Renewable Energy 103 (2017) 15e29 17

coefficient of determination. Ouarda et al. [22] estimate wind po- turbine investments. Knowing the mean annual wind speed is not
wer using three different probability density function: parametric enough to provide a precise energy calculation even if the mean
model, mixture model and non-parametric model with kernel annual wind velocity is determined on the basis of measurements
density concept. Results indicate that the Weibull distribution fits taken over decades [1]. It also requires information on how
better than three-parameter distributions used for comparison. frequently the individual wind speeds of the spectrum can be ex-
All cited papers confirm that the wind speed can be represented pected at a given location. To achieve a reliable statistical basis, it is
by a probability density function. The various studies indicate that necessary to take an evaluation period of at least several years [13].
the two-parameter Weibull distribution is the most popular model
to express the wind speed frequency distribution and to calculate 4. Wind speed distribution
the wind energy. On the other hand, although some advantages of
the Weibull distribution over the other distributions used in wind Wind speed distribution determines the wind energy available
energy field are shown, the two-parameter Weibull distribution and the performance of an energy conversion system for a partic-
does not reveal good conformity for the low wind speed. For that ular location. Wais [23] describes the wind evaluation technique
purpose, three-parameter Weibull distribution can appropriate and explains its relation to energy production. Knowledge of wind
correspond to that wind data and give better results in suitable characteristics helps to define site requirements, to choose a proper
conditions. turbine design based on wind turbine power characteristics and to
estimate the expected, average energy productivity having in mind
that the turbine output fluctuates. Measuring the wind on the site is
3. Theoretical, available wind power
the most reliable method. With a limited amount of wind data from
a given site (e.g. average wind speed only), the statistical method
Estimation of the available energy in the wind at a site, in which
can be used. It should be mentioned that statistically predictable
the wind turbine is proposed to be located, is one of the preliminary
values also require long-term measurements. The data, analysed in
steps in the planning of a wind energy project.
the article, is obtained from three weather stations:
The available power of the free-air stream that flows through
the cross-sectional area AR, at constant velocity v, is:
a) Site A, located in north-east part of Poland, measurement period
1 1 1 1971e2000,
PAVAIL v _ 2 ¼ rQ v2 ¼ rAR v3
¼ Pv ¼ mv (1) b) Site B, located in south-east part of Poland, measurement period
2 2 2
1985e2000,
where r is the density of air, Q is the volume flow passing through c) Site C, located in north-west part of Poland, measurement
the given cross section, AR ¼ pD2/4 is a rotor cross-sectional area, period 1971e2000.
where D is the turbine rotor diameter.
Eq. (1) demonstrates that the factors influencing the available To estimate the turbine productivity, it is necessary to know
power in the wind stream are the area of the wind rotor, the air wind velocity and the number of hours in the year, for which the
density, and the wind velocity. Effect of the wind velocity on the wind blows at velocity v. For three analysed meteorological sta-
wind power value is more prominent owing to its cubic relation- tions, wind measurements are available in time series format
ship. The determination of the wind conditions at the intended collected and sorted into classes. Then the set of data is separated
site of the wind turbine setting is an important task because into the wind speed intervals and appropriately. Similar to other
small changes in wind velocity can result in significant variation in wind research analysis [3e5,7e12,16,18,19,24e28], the size of wind
energy production. Because the wind velocity varies, it is necessary speed interval, used in the paper, is 1 m/s. Tables 1e3 present the
to know the frequency distribution and to give the information on direction and the yearly time in hours, in which the wind blows
the number of hours for which the velocity is within a specific with the given velocity.
range [14]. Then, the time given in Tables 1e3 can be easily converted into
Providing reliable wind data should be the first step in any wind frequency distribution format, in which the frequency of the wind

Table 1
Wind direction and wind velocity at Site A.

Wind speed Wind direction

m/s
Calm N NNE NEE E SEE SSE S SSW SWW W NWW NNW
h h h h h h h h h h h h h

0 746 13 5 2 1 2 2 0 5 11 14 8 5
1 0 84 130 150 118 75 74 44 78 105 92 79 97
2 0 90 146 160 119 136 140 122 144 151 156 142 84
3 0 73 111 102 109 88 99 113 166 233 162 151 124
4 0 63 65 57 69 94 128 85 134 165 160 135 87
5 0 38 47 23 40 45 129 87 80 102 190 131 69
6 0 18 15 13 24 30 46 27 50 104 117 125 69
7 0 10 6 9 7 16 40 22 33 63 111 73 42
8 0 5 1 0 1 10 25 22 16 44 56 59 23
9 0 2 0 0 0 2 5 12 20 21 36 27 19
10 0 0 0 0 0 0 0 0 0 0 0 0 0
11 0 0 0 0 0 0 0 0 0 0 0 0 0
12 0 0 0 0 0 0 0 0 0 0 0 0 0
13 0 0 0 0 0 0 0 0 0 0 0 0 0
14 0 0 0 0 0 0 0 0 0 0 0 0 0
15 0 0 0 0 0 0 0 0 0 0 0 0 0
Above 15 0 0 0 0 0 0 0 0 0 0 0 0 0
18 P. Wais / Renewable Energy 103 (2017) 15e29

Table 2
Wind direction and wind velocity at Site B.

Wind speed Wind direction

m/s
Calm N NNE NEE E SEE SSE S SSW SWW W NWW NNW
h h h h h h h h h h h h h

0 213 7 0 0 0 2 1 0 1 1 0 0 0
1 0 48 46 48 54 83 101 83 65 70 80 134 100
2 0 74 61 52 73 202 203 227 120 74 105 232 213
3 0 83 67 98 164 228 128 213 178 53 92 236 167
4 0 86 96 85 189 142 60 218 205 41 68 189 175
5 0 61 62 70 135 103 59 168 218 22 59 115 105
6 0 40 22 29 88 36 35 106 119 16 26 78 72
7 0 26 11 21 44 12 7 68 123 3 25 67 27
8 0 16 1 1 2 5 7 46 91 4 12 25 6
9 0 11 0 0 0 0 1 28 48 0 5 18 7
10 0 3 0 0 0 0 1 16 26 0 2 14 7
11 0 0 0 0 0 0 1 9 7 0 0 8 3
12 0 0 0 0 0 0 0 2 9 0 0 1 1
13 0 0 0 0 0 0 0 1 2 0 0 0 0
14 0 0 0 0 0 0 0 0 0 0 0 0 0
15 0 0 0 0 0 0 0 0 0 0 0 0 0
Above 15 0 0 0 0 0 0 0 0 1 0 0 0 0

Table 3
Wind direction and wind velocity at Site C.

Wind speed Wind direction

m/s
Calm N NNE NEE E SEE SSE S SSW SWW W NWW NNW
h h h h h h h h h h h h h

0 441 13 2 0 0 0 0 0 3 0 0 0 0
1 0 69 130 126 108 61 31 30 72 170 101 62 65
2 0 128 169 184 210 121 43 47 151 408 191 102 63
3 0 126 123 68 124 156 58 51 177 421 191 105 65
4 0 65 98 33 49 150 55 53 229 321 173 92 35
5 0 44 52 19 37 133 56 50 176 233 130 47 25
6 0 21 5 5 24 81 29 32 163 156 50 21 12
7 0 5 4 3 16 64 18 31 152 98 36 32 4
8 0 1 1 0 8 11 14 22 69 62 8 10 1
9 0 0 0 0 0 14 1 17 39 29 6 1 1
10 0 0 0 0 0 1 0 3 23 8 3 0 0
11 0 0 0 0 0 1 0 2 6 1 3 0 0
12 0 0 0 0 0 0 0 0 5 2 0 0 0
13 0 0 0 0 0 0 0 0 0 3 0 0 0
14 0 0 0 0 0 0 0 0 0 1 0 0 0
15 0 0 0 0 0 0 0 0 0 0 0 0 0
Above 15 0 0 0 0 0 0 0 0 0 0 0 0 0

Table 4
The wind velocity occurrence in Site A, B and C.

Velocity, Site A Site B Site C

vi, m/s
Hour in a year, Fraction of total time, Hour in a year, Fraction of total time, Hour in a year, Fraction of total time,
h p(vi) h p(vi) h p(vi)

0 814 0.09292 225 0.02568 459 0.05240

1 1126 0.12854 912 0.10411 1025 0.11701
2 1590 0.18151 1636 0.18676 1817 0.20742
3 1531 0.17477 1707 0.19486 1665 0.19007
4 1242 0.14178 1554 0.17740 1353 0.15445
5 981 0.11199 1177 0.13436 1002 0.11438
6 638 0.07283 667 0.07614 599 0.06838
7 432 0.04932 434 0.04954 463 0.05285
8 262 0.02991 216 0.02466 207 0.02363
9 144 0.01644 118 0.01347 108 0.01233
10 0 0.00000 69 0.00788 38 0.00434
11 0 0.00000 28 0.00320 13 0.00148
12 0 0.00000 13 0.00148 7 0.00080
13 0 0.00000 3 0.00034 3 0.00034
14 0 0.00000 0 0.00000 1 0.00011
15 0 0.00000 0 0.00000 0 0.00000
Above 15 0 0.00000 1 0.00011 0 0.00000
 P. Wais / Renewable Energy 103 (2017) 15e29 19

Table 5
Weibull parameters calculated by means of least square method.

2 parameter Weibull distribution 3 parameter Weibull distribution

Site A Site B Site C Site A Site B Site C

k 1.6545771 1.9474471 1.7959645 2.0974232 2.0752575 1.9665775

A 4.1953066 4.1879121 4.0197485 5.1039554 4.4113433 4.3453782
u e e e 1.2692170 0.2728700 0.4318956

speed that falls within various ranges is given. The total number of
Table 6
Coefficient of determination and fit standard error. hours in a year, tYEAR, is 8760. Table 4 shows the fraction of the time
at which wind speed is at v (for each bin of the wind speed).
2 parameter Weibull distribution 3 parameter Weibull distribution
Using the wind data, the wind speed vi and its frequency, the
Site A Site B Site C Site A Site B Site C available wind power and energy can be estimated by adding up
r2
0.8532976 0.9872793 0.9487926 0.9913127 0.9950736 0.9830247 the energy corresponding to all possible wind speeds. Because for a
SE 0.0247140 0.0079037 0.0156438 0.0059987 0.0049061 0.0089841 given velocity vi:

Fig. 1. Measurement data (points) and Weibull probability density curves for Site A.

Fig. 2. Measurement data (points) and Weibull probability density curves for Site B.
20 P. Wais / Renewable Energy 103 (2017) 15e29

Fig. 3. Measurement data (points) and Weibull probability density curves for Site C.

Table 7
Available wind power and energy.

Wind speed range: 0 m/seinfinity Wind speed range: 3.5 m/seinfinity

Site A Site B Site C Site Suw Site Kr Site SZCZ

PAVAIL_M, kW 429.639 512.269 449.658 399.374 479.307 416.423

PAVAIL_2P, kW 602.151 483.459 471.498 570.340 448.545 436.883
PAVAIL_3P, kW 437.681 457.353 425.710 407.415 422.778 391.781

EAVAIL_M, kWh 37,63,638 44,87,476 39,39,004 34,98,516 41,98,729 36,47,865

EAVAIL_2P, kWh 52,74,843 42,35,101 41,30,322 49,96,178 39,29,254 38,27,095
EAVAIL_3P, kWh 38,34,086 40,06,412 37,29,220 35,68,955 37,03,535 34,32,002

1 1 1 Xn
PAVAIL i ¼ _ 2i ¼ rAR $v3i
mv (2) PAVAIL ¼ rAR $ pðvi Þ$v3i (5)
2 2 M
2 i¼1
and Eq. (4) and Eq. (5) enable to calculate the available wind energy
and wind power directly from the measurements.
ti ¼ pðvi Þ$tYEAR (3)

then the available wind energy and power can be given by:

1 X
n
EAVAIL M ¼ rAR $tYEAR pðvi Þ$v3i (4) 5. Weibull probability distribution
2
i¼1
The statistical calculation of the wind energy requires the
or knowledge of the distribution of the wind speeds at a given site. A
correct determination of the probability distribution for wind
Table 8 speed values is very important in evaluating wind speed energy in a
Available wind power and energy within the limits of the cut-in and cut-out region.
velocities.

Wind speed range: 3.5 m/se25 m/s Table 9

Relative error of energy calculations.
Site A Site B Site C
Relative error, %
PAVAIL_M, kW 399.374 479.307 416.423
PAVAIL_2P, kW 570.340 448.545 436.883 Wind speed range: 0 m/s Wind speed range: 3.5 m/s
PAVAIL_3P, kW 407.415 422.778 391.781 einfinity e25 m/s

Site A Site B Site C Site A Site B Site C

EAVAIL_M, kWh 34,98,516 41,98,729 36,47,865
EAVAIL_2P, kWh 49,96,178 39,29,254 38,27,095 2 p Weibull distr 40.15 5.62 4.86 42.81 6.42 4.91
EAVAIL_3P, kWh 35,68,955 37,03,535 34,32,002 3 p Weibull distr 1.87 10.72 5.33 2.01 11.79 5.92
 P. Wais / Renewable Energy 103 (2017) 15e29 21

Table 10
The wind velocity occurrence in months M1, M2 and M3.

Velocity, M1 M2 M3
vi, m/s
Hour in a month, Fraction of total time, Hour in a month, Fraction of total time, Hour in a month, Fraction of total time,
h p(vi) h p(vi) h p(vi)

0 110.33 0.14830 68.33 0.09491 66.17 0.08895

1 135.33 0.18190 48.83 0.06782 46.17 0.06207
2 81.00 0.10887 53.17 0.07384 88.17 0.11853
3 82.17 0.11044 81.17 0.11273 121.33 0.16312
4 102.83 0.13822 116.33 0.16157 123.67 0.16626
5 93.00 0.12500 123.33 0.17130 93.67 0.12592
6 63.33 0.08513 102.83 0.14282 96.33 0.12951
7 33.17 0.04458 67.00 0.09306 59.00 0.07932
8 24.00 0.03226 32.00 0.04444 25.83 0.03473
9 13.67 0.01837 18.33 0.02546 16.67 0.02241
10 4.00 0.00538 7.83 0.01088 4.17 0.00560
11 0.67 0.00090 0.83 0.00116 1.00 0.00134
12 0.17 0.00022 0.00 0.00000 1.00 0.00134
13 0.33 0.00045 0.00 0.00000 0.67 0.00090
14 0.00 0.00000 0.00 0.00000 0.00 0.00000
15 0.00 0.00000 0.00 0.00000 0.00 0.00000
Above 15 0.00 0.00000 0.00 0.00000 0.00 0.00000

Table 11
The wind velocity occurrence in months M4, M5 and M6.

Velocity, M4 M5 M6
vi, m/s
Hour in a month, Fraction of total time, Hour in a month, Fraction of total time, Hour in a month, Fraction of total time,
h p(vi) h p(vi) h p(vi)

0 53.00 0.07124 282.17 0.37926 71.83 0.09977

1 35.00 0.04704 186.33 0.25045 42.67 0.05926
2 59.00 0.07930 91.00 0.12231 43.67 0.06065
3 123.17 0.16555 66.83 0.08983 62.67 0.08704
4 148.67 0.19982 40.50 0.05444 92.33 0.12824
5 140.83 0.18929 30.33 0.04077 112.33 0.15602
6 90.83 0.12209 22.50 0.03024 108.17 0.15023
7 52.33 0.07034 12.33 0.01658 77.50 0.10764
8 23.17 0.03114 9.50 0.01277 50.33 0.06991
9 10.50 0.01411 1.83 0.00246 31.17 0.04329
10 4.50 0.00605 0.50 0.00067 18.83 0.02616
11 1.33 0.00179 0.00 0.00000 7.50 0.01042
12 1.00 0.00134 0.17 0.00022 1.00 0.00139
13 0.33 0.00045 0.00 0.00000 0.00 0.00000
14 0.00 0.00000 0.00 0.00000 0.00 0.00000
15 0.17 0.00022 0.00 0.00000 0.00 0.00000
Above 15 0.17 0.00022 0.00 0.00000 0.00 0.00000

Table 12
The wind velocity occurrence in months M7, M8 and M9.

Velocity, M7 M8 M9
vi, m/s
Hour in a month, Fraction of total time, Hour in a month, Fraction of total time, Hour in a month, Fraction of total time,
h p(vi) h p(vi) h p(vi)

0 56.17 0.07549 3.17 0.00440 16.33 0.02347

1 46.33 0.06228 42.33 0.05880 26.33 0.03784
2 94.67 0.12724 76.17 0.10579 41.17 0.05915
3 161.50 0.21707 100.83 0.14005 49.67 0.07136
4 170.67 0.22939 116.83 0.16227 61.50 0.08836
5 105.33 0.14158 109.67 0.15231 100.50 0.14440
6 61.67 0.08289 82.67 0.11481 101.50 0.14583
7 28.33 0.03808 74.67 0.10370 92.50 0.13290
8 10.67 0.01434 51.33 0.07130 55.50 0.07974
9 5.67 0.00762 29.33 0.04074 41.00 0.05891
10 1.50 0.00202 19.33 0.02685 37.83 0.05436
11 1.00 0.00134 8.50 0.01181 32.00 0.04598
12 0.33 0.00045 3.17 0.00440 19.17 0.02754
13 0.00 0.00000 1.83 0.00255 11.17 0.01604
14 0.17 0.00022 0.17 0.00023 5.17 0.00742
15 0.00 0.00000 0.00 0.00000 3.83 0.00551
Above 15 0.00 0.00000 0.00 0.00000 0.83 0.00120
22 P. Wais / Renewable Energy 103 (2017) 15e29

Table 13
Weibull parameters calculated by means of least square method for M1, M2 and M3.

2 parameter Weibull distribution 3 parameter Weibull distribution

M1 M2 M3 M1 M2 M3

k 1.2558070 2.3657442 2.0593091 1.3985159 87.63695 4.1012388

A 4.5899040 5.6126697 5.1047404 4.6830761 204.04873 10.0009200
u e e e 0.7997338 199.06991 5.3583155

Table 14
Weibull parameters calculated by means of least square method for M4, M5 and M6.

2 parameter Weibull distribution 3 parameter Weibull distribution

M4 M5 M6 M4 M5 M6

k 2.5950924 1.0001481 2.3481882 7.2934806 0.99450 22.1332914

A 5.0505128 2.6094945 6.2616936 13.9549047 2.06874 58.6781041
u e e e 9.2067150 0.48788 53.1484891

Table 15
Weibull parameters calculated by means of least square method for M7, M8 and M9.

2 parameter Weibull distribution 3 parameter Weibull distribution

M7 M8 M9 M7 M8 M9

k 2.4818007 2.0820993 2.3135340 123.6314412 2.13005 2.9769205

A 4.3367656 5.6247757 7.1987878 210.2758797 5.72550 8.9757595
u e e e 206.3572630 0.11551 1.9728496

5.1. Two-parameter Weibull distribution

Z∞
The two-parameter Weibull probability density function is PAVAIL ¼ Pv $pðvÞ dv (7)
expressed by: 0

Introducing Eq. (1) and Eq. (6), Eq. (7) is written as:

  k1  k  k
k v  v    k1 Z∞ v
pðvÞ ¼
A 1 k 1 k1  A
A A
e (6) PAVAIL 2P ¼ rAR $ $ 3
v $ðvÞ e dv (8)
2 A A
0
where v is the wind speed and v > 0, k > 0, A > 0.
Dimensionless factor k determines the shape of the curve and is or
called a shape factor. Parameter A, in m/s, is the scale parameter.  k
The distributions take different shapes with different values of k Z∞ v
1 1 kþ2 
¼ r$AR $k$ k ðvÞ
A
and A. PAVAIL 2P e dv (9)
2 A
The energy available in the wind may also be estimated using 0
the Weibull probability density function. Once the wind velocities
and its distribution are available, then knowing the probability p (v) and
for which the wind velocity is v, the total available wind power can
be written as: EAVAIL 2P ¼ PAVAIL 2P $tYEAR (10)
Eq. (9) and Eq. (10) allow to calculate the available wind power
Table 16 and energy for two-parameter Weibull distribution.
Coefficient of determination and fit standard error for M1, M2 and M3.

2 parameter Weibull distribution 3 parameter Weibull distribution

5.2. Three-parameter Weibull distribution
M1 M2 M3 M1 M2 M3

r2 0.5046128 0.7688464 0.9095352 0.8829550 0.9257131 0.9223213 The three-parameter Weibull probability density function is
SE 0.0435647 0.0284898 0.0250835 0.0211217 0.0161096 0.0231840
given by Eq. (11):

Table 17 Table 18
Coefficient of determination and fit standard error for M4, M5 and M6. Coefficient of determination and fit standard error for M7, M8 and M9.

2 parameter Weibull distribution 3 parameter Weibull distribution 2 parameter Weibull distribution 3 parameter Weibull distribution

M4 M5 M6 M4 M5 M6 M7 M8 M9 M7 M8 M9
2 2
r 0.9037604 0.9633553 0.6916422 0.9505583 0.9934738 0.8746231 r 0.9069524 0.9939454 0.8887179 0.9457762 0.9940170 0.9022570
SE 0.0210574 0.0194379 0.0295040 0.0150544 0.0081820 0.0187651 SE 0.0227456 0.0044691 0.0151458 0.0173192 0.0044313 0.0141584
 P. Wais / Renewable Energy 103 (2017) 15e29 23

Fig. 4. Measurement data (points) and Weibull probability density curves for M1.

 k
   vu
 k
k v  u k1  A
pðvÞ ¼ e (11) Z∞  vu
A A 1 1 A
PAVAIL ¼ rA $k$ k v3 $ðv  uÞk1 e dv (13)
3P
2 R A
where factor k, no unit of measurement, is the Weibull shape 0
parameter and parameter A is the scale parameter in a unit equal to
the wind speed unit. The third parameter, u, is a location parameter, and
and its value is lower than zero. The Weibull location parameter
EAVAIL 3P ¼ PAVAIL 3P $tYEAR (14)
unit is the same as the parameter A.
Introducing Eq. (1) and Eq. (11), Eq. (7) is written as: Eq. (13) and Eq. (14) enables to calculate the available wind
power and energy for three-parameter Weibull distribution.
 k
   k1 Z∞  vu
1 k 1 A
PAVAIL 3P ¼ rAR $ $ v3 $ðv  uÞk1 e dv 6. Statistical analysis of the wind measurements
2 A A
0
or Three statistical wind data sets are analysed using Weibull
(12)

Fig. 5. Measurement data (points) and Weibull probability density curves for M2.
24 P. Wais / Renewable Energy 103 (2017) 15e29

Fig. 6. Measurement data (points) and Weibull probability density curves for M3.

distributions in order to investigate the Weibull parameters,

shape and scale parameters for two-parameter Weibull distri- X
n
bution and shape, scale and location parameters for three- SSE ¼ ½pðvi Þ  f ðvi Þ2 /minimum (15)
parameter Weibull distribution. Considering the literature i¼1
[2,7,11,26,29,30], the least square method is one of the most In Eq. (15), vi is the wind speed, p (vi) is the probability of the
common technique to calculate the parameters in a Weibull wind speed received from the measurements, f (vi) is an estimated
formula for a wind resource assessment. Therefore, the Weibull value of the probability for the wind speed vi.
parameters are computed using the least square method in the Table Curve 2D computer program is used in order to evaluate
paper. The unknown parameters of a Weibull probability density the parameters of the Weibull probability distribution functions
function are estimated by looking for the numerical values of [31]. The internal algorithm, applied to determine parameters and
those parameters, for which the sum of the squares of the de- to estimate the degree of fit, is based on the least square method.
viations between the measured values and those obtained from Calculated Weibull parameters are shown in Table 5.
the density function is minimised. At least square method, the The coefficient of determination r2, and the fit standard error SE
sum of the squares due to error SSE (sum of residual squared) is are considered to check the correctness of parameter calculations.
given by: Table 6 shows the results for the theoretical distributions and the

Fig. 7. Measurement data (points) and Weibull probability density curves for M4.
 P. Wais / Renewable Energy 103 (2017) 15e29 25

Fig. 8. Measurement data (points) and Weibull probability density curves for M5.

corresponding measured frequencies. three locations.

Analysing the coefficient in Table 6, it can be noticed that the Available wind power and energy in three locations are calcu-
three-parameter Weibull distribution provides a better fit lated using three methods:
comparing to two-parameter Weibull distribution for all three
stations analysed in the paper. 1) value received directly from measurements, Eq. (4) and (5)
2) value computed from two-parameter Weibull distribution, Eq.
7. Results (9) and (10),
3) value received from three-parameter Weibull distribution, Eq.
The model accuracy is not based only on the coefficient of (13) and (14).
determination and the fit standard error parameters. The suitability
of the two and three-parameter Weibull distribution is also judged The frequency of wind velocity measurements and Weibull
from the calculated energy production. Comparisons are carried probability density curves are given in Figs. 1e3.
out between each available wind energy obtained from the distri- Looking at Figs. 1e3 it can be seen that the significant discrep-
bution and the wind energy gathered from the data analysis for ancy appears at site A, for which the frequency of null wind is

Fig. 9. Measurement data (points) and Weibull probability density curves for M6.
26 P. Wais / Renewable Energy 103 (2017) 15e29

Fig. 10. Measurement data (points) and Weibull probability density curves for M7.

higher. It can be noticed that for other two meteorological stations, speeds, at which the turbine starts producing the energy, Table 8.
the difference between two and three-parameter Weibull distri- Three real data sets are used to verify the Weibull distributions.
butions is not so essential. The results in Tables 7 and 8 show that the annual available wind
To calculate the available wind energy, it is necessary to take energy, evaluated by means of Weibull distribution, can be over- or
some assumptions and to know the air density (presumed as underestimated relating to the results received from measure-
r ¼ 1.225 kg/m3) and the rotor diameter (presumed as D ¼ 100 m). ments (PAVAIL_M and EAVAIL_M). Table 9 presents the difference be-
For measurement data, the power can be computed by adding up tween the energy obtained from the data and the energy gained
the energy corresponding to all possible wind speeds. For Weibull from Weibull distributions in relation to value EAVAIL_M calculated
distributions, all the computation procedures are implemented in from the wind velocity measurements.
Matlab software package, in which Eq. (9), Eq. (10), Eq. (13) and Eq. It is observed that the three-parameter Weibull distribution
(14) are computed. Available wind power and wind energy are gives better results of energy calculation when the frequency of the
expressed in Table 7. null wind is higher (Site A, in which the frequency of the null wind
Additionally, because most of the wind turbine cut-in velocity is is 9.29%). Nevertheless, the two-parameter Weibull distribution
about 3.5 m/s, the available wind power is calculated for the wind provides a better fit for other two sites (the frequency of the null

Fig. 11. Measurement data (points) and Weibull probability density curves for M8.
 P. Wais / Renewable Energy 103 (2017) 15e29 27

Fig. 12. Measurement data (points) and Weibull probability density curves for M9.

wind in Site B is 2.57%, the frequency of null wind in Site C is 5.24%). Weibull parameters, calculated for new sets of data by means of
least square method, are presented in Tables 13e15.
8. Discussion The coefficient of determination r2, and the fit standard error SE
are calculated to check the correctness of distribution parameter
The reason why 2-parameter distribution performs better at two calculations. The received values are presented in Tables 16e18.
sites are further investigated and additional research is conducted It is observed that also in these additional analysed cases, the
to confirm observed results. A new set of data is examined for three-parameter Weibull distributions fit better the sets of wind
monthly periods. That approach gives the possibility to research velocity data. The Weibull density curves are illustrated in
several supplementary Weibull distributions, calculate the coeffi- Figs. 4e12.
cient of determination and the fit standard error as well as the wind Analysing the behaviour of curves that represent density func-
available energy and wind power. For wind energy applications, tions, it can be noticed that the three-parameter Weibull distribu-
wind speed is typically measured every 1e3 s and 10-min statistics tion is above the two-parameter Weibull distribution for smaller
like average, minimum, maximum, and standard deviation are values of wind velocity, except for Weibull probability density
computed and recorded [32]. The mean wind velocity should be curves for M5. On the contrary, the two-parameter Weibull distri-
based on the longest practical interval that can be regarded as bution gives higher values of the distribution probability for higher
stationary. In practice, the 10-min average generally satisfies this wind velocity and the two-parameter Weibull distribution is above
requirement and World Meteorological Organization standard for three-parameter Weibull distribution. The three-parameter Wei-
estimating the mean wind is the 10-min average [33]. Then, the bull distribution function fits the curve better, but because the
wind speed mean value for 10 min is evaluated for monthly mea- available wind energy and wind power are wind cubic velocity
surements analysed in the paper. Tables 10e12 presents the wind functions, the two-parameter Weibull distribution can give higher
velocity and its fraction of time.

Table 21
Table 19
Available wind power and energy production for M7, M8 and M9.
Available wind power and energy production for M1, M2 and M3.
Wind speed range: 0 m/seinfinity
Wind speed range: 0 m/seinfinity
M7 M8 M9
M1 M2 M3
PAVAIL_M, kW 425.117 1042.810 2053.210
PAVAIL_M, kW 461.796 1305.379 713.209
PAVAIL_2P, kW 434.397 1092.602 2089.698
PAVAIL_2P, kW 1366.214 1529.467 973.813
PAVAIL_3P, kW 281.539 1075.399 1830.596
PAVAIL_3P, kW 850.485 1305.219 598.008

Table 22
Table 20
Relative error of energy calculations for M1, M2 and M3.
Available wind power and energy production for M4, M5 and M6.
Relative error, %
Wind speed range: 0 m/seinfinity
Wind speed range: 0 m/seinfinity
M4 M5 M6
M1 M2 M3
PAVAIL_M, kW 598.132 148.946 1013.164
PAVAIL_2P, kW 667.044 510.874 1359.728 2 p Weibull distr 195.85 17.17 36.54
PAVAIL_3P, kW 509.420 206.567 840.391 3 p Weibull distr 84.17 0.01 16.15
28 P. Wais / Renewable Energy 103 (2017) 15e29

Table 23 two Sites B and C are further investigated. New sets of data are
Relative error of energy calculations for M4, M5 and M6. examined for monthly periods. That approach gives the possibility
Relative error, % to research several supplementary Weibull distributions.
Wind speed range: 0 m/seinfinity
The general conclusion is that the two-parameter Weibull dis-
tribution is more suitable when the frequency of null velocity is
M4 M5 M6
insignificant and for a higher percentage of elevated wind speeds.
2 p Weibull distr 11.52 242.99 34.21 Otherwise, the three-parameter Weibull distribution can offer
3 p Weibull distr 14.83 38.69 17.05
minor relative errors in determining the annual energy production.
The three-parameter Weibull distribution takes into account the
frequency of null winds and it can represent a useful alternative to
values of energy production comparing to the three-parameter the two-parameter Weibull distribution for the wind with consid-
Weibull distribution. The available wind power and energy are erable null wind probability and a higher fraction of lower speed
presented in Tables 19e21. winds. The investigation shows that the third parameter in the
Tables 22e24 present the difference between the energy ob- Weibull distribution, the location parameter, can help to take the
tained from the data and the energy gained from Weibull distri- right decision and select the site for the wind turbine plant. How-
butions in relation to value EAVAIL_M calculated from the wind ever, fitting a three-parameter Weibull, where the third parameter
velocity measurements. is a location parameter, can lead to the negative values of the
Evaluating the results, it can be noticed that the two-parameter probability density function. In such cases, the lower limit of the
Weibull distribution is a useful tool for wind energy estimation distribution should be controlled to avoid reaching unphysical
although it can sometimes be inadequate to model wind distribu- negative wind speeds. It should be also noticed that density func-
tion and can give wrong results for the wind speed data with higher tion is only one of the factors in the energy equations. Wind ve-
frequency of the null wind. Under such circumstances, the three- locity influences the energy calculation in a more significant way.
parameter Weibull distribution, according to the results of the Because power available in wind increases as the cube of velocity,
studies, may be suitable for wind energy calculations and give more and the two-parameter Weibull distributions can reach higher
precise outcomes. value comparing to three-parameter Weibull distribution for
The coefficient of determination and the fit standard error can greater wind velocities, the energy gained from two-parameter
check the correctness of distribution parameter calculations. Weibull distribution can be higher and better match the calcu-
Comparing these two parameters between two parameter and lated energy from velocity measurements. These circumstances do
three-parameter Weibull distribution for the same set of data, it can not mean that the three-parameter Weibull distribution is less
be noticed that the three-parameter Weibull distribution better fits accurate or fits improperly the probability curves.
to the input data in all evaluated cases. The reason why the two
parameter distribution can give closer results to energy outcome Acknowledgement
calculated from the wind velocity measurements is related to the
wind cubic velocity functions in Equations (8) and (12). This research did not receive any specific grant from funding
agencies in the public, commercial, or not-for-profit sectors.
9. Conclusions
References
In the paper, two diverse Weibull distributions are presented to
estimate the theoretical, available wind power. A three-parameter [1] P. Wais, Hub height and wind turbine power plant payback period (original
title: wysokosc wiezy a okres zwrotu nakladow w energetyce wiatrowej),
Weibull distribution is proposed for wind energy calculation as
Rynek Energii 4 (2013) 23e32.
an alternative for two-parameter Weibull distribution. Suitable [2] M. Jamil, S. Parsa, M. Majidi, Wind power statistics and an evaluation of wind
examples, based on the three real data sets, are provided to energy density, Renew. Energy 6 (1995) 623e628.
examine the properties of two different Weibull distributions. Two [3] I.Y.F. Lun, J.C. Lam, A study of Weibull parameters using long-term wind ob-
servations, Renew. Energy 20 (2000) 145e153.
appropriate probability density functions are studied, compared [4] J.V. Seguro, T.W. Lambert, Modern estimation of the parameters of the Weibull
and correlated to the wind measurements at three different sites in wind speed distribution for wind energy analysis, J. Wind Eng. Ind. Aerodyn.
Poland. The results, received from distribution analyses, are 85 (2000) 75e84.
[5] A.N. Celik, A statistical analysis of wind power density based on the Weibull
compared to energy outcome gathered from a real data set. The and Rayleigh models at the southern region of Turkey, Renew. Energy 29
paper presents that the two-parameter Weibull distribution is not (2003) 593e604.
always sufficient to specify the wind speed distribution and to [6] A.S.S. Dorvlo, Estimating wind speed distribution, Energy Convers. Manag. 43
(2002) 2311e2318.
evaluate the available wind power. In addition, it is observed that [7] J.A. Carta, P. Ramirez, Analysis of two-component mixture Weibull statistics
the three-parameter Weibull distribution can offer the advantage for estimation of wind speed distributions, Renew. Energy 32 (2007)
for some wind data analyses and fits certain data better than the 518e531.
[8] P. Ramirez, J.A. Carta, The use of wind probability distributions derived from
typical Weibull distribution, especially for the wind with a signifi- the maximum entropy principle in the analysis of wind energy. A case study,
cant percentage of null wind speed. Energy Convers. Manag. 47 (2006) 2564e2577.
The reason why two parameter distribution performs better at [9] S.A. Akdag, H.S. Bagiorgas, G. Mihalakakou, Use of two-component Weibull
mixtures in the analysis of wind speed in the Eastern Mediterranean, Appl.
Energy 87 (2010) 2566e2573.
[10] P.A.C. Rocha, R. Coelho de Sousa, C. Freitas de Andrade, M.E. Vieira da Silva,
Table 24
Comparison of seven numerical methods for determining Weibull parameters
Relative error of energy calculations for M7, M8 and M9. for wind energy generation in the northeast region of Brazil, Appl. Energy 89
(2012) 395e400.
Relative error, %
[11] J.A. Carta, P. Ramirez, S. Velazquez, A review of wind speed probability dis-
Wind speed range: 0 m/seinfinity tributions used in wind energy analysis Case studies in the Canary Islands,
Renew. Sustain. Energy Rev. 13 (2009) 933e955.
M7 M8 M9 [12] T.C. Carneiro, S.P. Melo, P.C.M. Carvalho, A.P.S. Braga, Particle Swarm Opti-
2 p Weibull distr 2.18 4.77 1.78 mization method for estimation of Weibull parameters: a case study for the
Brazilian northeast region, Renew. Energy 86 (2016) 751e759.
3 p Weibull distr 33.77 3.13 10.84
[13] E. Hau, Wind Turbines. Fundamentals, Technologies, Application, Economics,
 P. Wais / Renewable Energy 103 (2017) 15e29 29

second ed., Springer, Berlin, 2006. U.C. Sharma, R. Prasad (Eds.), Energy Science and Technology. Wind Energy,
[14] S. Mathew, Wind Energy. Fundamentals, Resource Analysis and Economics, Studium Press LLC, Houston, USA, 2015, pp. 168e193.
Springer, Berlin, 2006. [24] A. Garcia, J.L. Torres, E. Prieto, A. De Francisco, Fitting wind speed distribu-
[15] IEC 61400e12, (E) Wind Turbine Generator Systems e Part 12: Wind Turbine tions: a case study, Sol. Energy 62 (1998) 139e144.
Power, 1998. [25] A.N. Celik, Energy output estimation for small-scale wind power generators
[16] T. Arslan, Y.M. Bulut, A.A. Yavuz, Comparative study of numerical methods for using Weibull-representative wind data, J. Wind Eng. Ind. Aerodyn. 91 (2003)
determining Weibull parameters for wind energy potential, Renew. Sustain. 693e707.
Energy Rev. 40 (2014) 820e825. [26] S. Akpinar, E.K. Akpinar, Estimation of wind energy potential using finite
[17] B. Yaniktepe, T. Koroglu, M.M. Savrun, Investigation of wind characteristics mixture distribution models, Energy Convers. Manag. 50 (2009) 877e884.
and wind energy potential in Osmaniye, Turkey, Renew. Sustain. Energy Rev. [27] S.A. Akdag, A. Dinler, A new method to estimate Weibull parameters for wind
21 (2013) 703e711. energy applications, Energy Convers. Manag. 50 (2009) 1761e1766.
[18] P. Ramirez, J.A. Carta, Influence of the data sampling interval in the estimation [28] C.F. Andrade, H.F.M. Neto, P.A. Costa Rocha, M.E. Vieira da Silva, An efficiency
of the parameters of the Weibull wind speed probability density distribution: comparison of numerical methods for determining Weibull parameters for
a case study, Energy Convers. Manag. 46 (2005) 2419e2438. wind energy applications: a new approach applied to the northeast region of
[19] T.P. Chang, Estimation of wind energy potential using different probability Brazil, Energy Convers. Manag. 86 (2014) 801e808.
density functions, Appl. Energy 88 (2011) 1848e1856. [29] K.E. Ahmad, Modified weighted least-squares estimators for the three-
[20] X. Qin, J.S. Zhang, X.D. Yan, Two improved mixture Weibull models for the parameter Weibull distribution, Appl. Math. Lett. 7 (1994) 53e56.
analysis of wind speed data, J. Appl. Meteorol. Climatol. 51 (2012) 1321e1332. [30] J.Y. Shin, T.B.M.J. Ouarda, T. Lee, Heterogeneous mixture distributions for
[21] S.H. Pishgar-Komleh, A. Keyhani, P. Sefeedpari, Wind speed and power den- modeling wind speed, application to the UAE, Renew. Energy 91 (2016)
sity analysis based on Weibull and Rayleigh distributions (a case study: Fir- 40e52.
ouzkooh country of Iran), Renew. Sustain. Energy Rev. 42 (2015) 313e322. [31] TableCurve 2D v5.01 for Windows©, SYSTAT Software Inc.
[22] T.B.M.J. Ouarda, C. Charron, J.-Y. Shin, P.R. Marpu, A.H. Al-Mandoos, M.H. Al- [32] J. Pramod, Wind Energy Engineering, McGraw eHill, New York, 2010.
Tamimi, H. Ghedira, T.N. Al Hosary, Probability distributions of wind speed in [33] Y. Tamura, A. Kareem, Advanced Structural Wind Engineering, Springer, Japan,
the UAE, Energy Convers. Manag. 93 (2015) 414e434. 2013.
[23] P. Wais, Wind energy and wind turbine selection, in: S. Sivakumar,