Energy 106 (2016) 301e314

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

An innovative estimation method regarding Weibull parameters for

wind energy applications
Ilhan Usta*
Department of Statistics, Faculty of Science, Anadolu University, Eskisehir 26470, Turkey

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

Article history: Two-parameter Weibull distribution is widely-used and accepted as the most popular distribution for
Received 1 July 2015 the evaluation of wind energy potential. However, it is well-known that parameter estimates have sig-
Received in revised form nificant effects on the success of Weibull distribution for wind energy applications. Thus, in this paper, an
2 February 2016
innovative method, PWMBP (probability weighted moments based on the power density method), is
Accepted 15 March 2016
developed and proposed for estimating the Weibull parameters in wind energy applications. The salient
features of PWMBP compared to other common methods are presented and discussed. In particular, this
method is expressed in analytical form and does not need extensive numerical calculations or any
Keywords:
Weibull distribution
iterative procedure. Moreover, the performance of PWMBP is compared with six other commonly-used
Parameter estimation methods; the maximum likelihood, modified maximum likelihood, graphical, moment, power density
Probability weighted moments method and probability weighted moments methods for actual wind data, based on different time periods and
Probability weighted moments based on the regions according to various goodness of fit criteria, moreover energy output performance of these
power density method methods are examined. The obtained results indicate that PWMBP provides more accurate and efficient
Wind energy estimation than other methods in estimating the parameters of Weibull distribution. As a result, PWMBP
can be used as an improved method to estimate the Weibull parameters for wind energy applications.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction The parameters of Weibull distribution are used to obtain ac-

curate results regarding wind energy potential and characteristics.
With a rise in consciousness of renewable and clean energy Thus, the various estimation methods for Weibull parameters have
sources, wind energy is being widely-used and has become more been proposed and compared among themselves [19e34]. For
popular in many countries in recent decades. Hence, an accurate example, Justus and Mikhail [21] proposed the MOM (moment
assessment of wind energy potential and characteristics for a method) to estimate the Weibull parameters when the mean and
certain location is of great importance. Researchers, engineers, and variance of wind speed are obtainable. Justus et al. [22] presented
designers agree that the probability distribution of wind speed is the four different estimation method and the GM (graphical
key information necessary to determine wind energy potential at a method) based on the concept of least squares method and also
particular location. For this reason, various pdf (probability density compared them. Stevens and Smulders [23] suggested the MLM
functions) have been used to describe wind speed distribution, (maximum likelihood method) for the estimation of the parameters
such as lognormal, gamma, Weibull, Rayleigh, logistic, maximum of Weibull wind speed distribution. Seguro and Lambert [24]
entropy principle-type and mixture distributions [1e9]. However, compared the commonly-used (MLM), (GM) and the proposed
two-parameter Weibull distribution is the most widely-used and MMLM (modified maximum likelihood method). As a result, they
accepted distribution in the prediction of wind energy potential in concluded that MLM performs better than GM. Dorvlo [25] esti-
the literature, considering its flexibility and simplicity, as well as it mated the Weibull parameters used to model wind speeds in Oman
being a good fit to wind data [10e18]. using three methods, the Chi-square method, MOM and GM. The
results showed that the Chi-square method gives better estimates
for Weibull parameters than the other methods. Silva [26] proposed
the EEM (equivalent energy method) that obtained by equating the
energy density of theoretical curve to the energy density of obser-
* Tel.: þ90 222 335 05 80x4680. vations. Jowder [27] aimed to compare the EM (empirical method)
E-mail address: [email protected].

http://dx.doi.org/10.1016/j.energy.2016.03.068
0360-5442/© 2016 Elsevier Ltd. All rights reserved.
302 I. Usta / Energy 106 (2016) 301e314

Nomenclature PDM power density method

PWM1,0,s probability weighted moments of a random variable X
A wind turbine rotor swept area (m2) PWMBP probability weighted moments based on power
c scale parameter of Weibull distribution (m/s) density method
Cp,eq coefficient of cubic approximation PWMM probability weighted moments method
cdf cumulative distribution function PREF mean power density based on time series wind data
EM empirical method (W/m2)
Epf energy pattern factor PWD mean power density based on Weibull distribution (W/
EREF energy production of a wind tribune based on time m2 )
series wind data (kWh) q(v) non-linear part of power curve
EWD energy production of a wind tribune based on Weibull R2 coefficient of determination
distribution (kWh) RMSE root mean square error
fvi relative frequency according to each wind speed vi S sample standard deviation of wind speed
f(vi) frequency for wind speed ranging within bin i v wind speed (m/s)
F(v) cumulative distribution function of Weibull vci cut-in wind speed (m/s)
distribution vci cut-out wind speed (m/s)
GM graphical method vi ith observed wind speed data (m/s)
k shape parameter of Weibull distribution v(i) ith ascending ordered wind speed data
KS KolmogoroveSmirnov distance vr rated wind speed (m/s)
LMOM L-moment method v3 sample mean of wind speed cubes
MLM maximum likelihood method v sample mean of wind speed
MMLM modified maximum likelihood method xj jth the observed probability of wind speed data
MOM moment method yj jth the predicted probability calculated from Weibull
n number of all observed non-zero wind speed data distribution
N number of bins y average of yj values
Nv number of observation corresponding to power curve
p shape parameter of Weibull distribution Greek letters
Pr rated output power G() Gamma function
P(v) energy production from a wind tribune (W) r air density (kg/m3)
pdf probability density function
PDE power density error

and GM for prediction of the average wind speed and power den- They also found that when the sample size is greater than 100, MLM
sity, using wind data measured in the Kingdom of Bahrain. The is preferable in comparison to other methods for the estimation of
results indicated that EM provides more accurate prediction than shape parameter in terms of the MSE (mean square error) criteria.
GM. Akdag and Dinler [28] used three conventional methods, In addition, Akdag and Dinler [34] presented a novel method to
namely GM, MLM and MOM to compare the proposed PDM (power estimate Weibull parameters. The proposed method was also
density method) for estimating Weibull parameters. Their results compared with the standard methods: GM, MLM, MMLM, MOM,
show that the PDM has better suitability than others according to EM, PDM, the alternative maximum likelihood and the WAsP
power density and mean wind speed. Furthermore, Chang [29] methods. The results provided that the novel method is adequate to
conducted a comparative study to show the performance of GM, determine Weibull distribution parameters.
MLM, MMLM, MOM, PDM and EM, in estimating Weibull parame- In a literature review, it is seen that there are studies focusing on
ters for wind energy application. Rocha et al. [30] studied the same Weibull parameters to predict wind resource accurately, yet there is
six methods as Chang [29] including one additional method; EEM. no consensus on the most accurate estimation. However, it is
They aimed to analyze and compare these seven numerical commonly agreed that parameter estimates have significant effects
methods for assessing effectiveness in determining parameters of on the success of Weibull distribution in wind energy applications
Weibull, using wind data collected in the northeast region of Brazil. to predict wind resource. Considering the need for new estimation
Along the same lines, Azad et al. [31] used the seven methods methods, this study aims to develop an innovative method called as
applied by Rocha et al. [30] to estimate Weibull parameters and the PWMBP (probability weighted moments based on power
used six statistical tools to rank the methods precisely. They found density method) to estimate more accurate parameters of Weibull
that MOM and MLM are the most efficient methods for estimating distribution for wind energy applications. The key issue in this
parameters of Weibull distribution. de Andrade et al. [32] also method is wind power that has come to the forefront as a crucial
performed a statistical analysis of seven mathematical methods element in predicting the feasibility, suitability and capability of a
used by Refs. [30,31] to estimate the shape and the scaling pa- given location for the effective use of wind energy [28,32,35]. How
rameters of Weibull, using wind speed and wind power data the design and application of this innovative method are explained
collected in the two coastal cities of northeast region of Brazil. Their in this paper through comparison with other methods in the
results indicated that the values generated for the shape and scale literature regarding energy in the following manner. Firstly, Section
parameters provide differences considering the databases formed 2 reviews GM, MLM, MMLM, MOM and PDM. The PWMM (proba-
by wind speed and wind power. In addition, Arslan et al. [33] bility weighted moments method) and PWMBP are in detail
compared MOM, ML and the LMOM (L-moment method) for esti- introduced in Section 3. In section 4, various goodness of fit criteria,
mation of wind speed parameters relevant to Weibull distribution. such as the coefficient of determination (R2), the RMSE (root mean
 I. Usta / Energy 106 (2016) 301e314 303

square error), the KS (KolmogoroveSmirnov) statistic and the PDE shape parameter k, at the second stage, the scale parameter c is
(power density error) are given to show the suitability and accuracy estimated from the following equation:
of PWMBP. The performance and accuracy of the proposed PWMBP
!1=k
over six other methods is compared in terms of the considered 1Xn
criteria as well as in terms of graphical illustration for actual wind c¼ vki (4)
n
data based on different time periods and regions of Turkey in i¼1

Section 5. Moreover, in Section 5, the comparison of the proposed

PWMBP and six other methods is conducted with regard to energy
production from a wind turbine which is calculated by using the
probability distribution and wind power curve. Section 6 concludes 2.2. The modified maximum likelihood method-MMLM
the study emphasizing the superiority and accuracy of PWMBP.
MMLM can be applied if wind speed data is available in fre-
quency distribution format [24]. Similar to MLM, the shape and
2. Review of methods for estimating Weibull parameters scale parameters of Weibull distribution are iteratively estimated
using the following two equations [24,30]:
The two-parameter Weibull distribution has been widely-used
and accepted as providing wind speed distribution and evaluating 0 11
wind energy potential in the literature on wind energy. The Weibull P
B N vk lnðv Þf ðv Þ P
B i i
N
lnðvi Þf ðvi ÞC
C
distribution can be expressed as a pdf (probability density function) k ¼ B i¼1
PN k
i
 i¼1 C (5)
f(v) and a cdf (cumulative distribution function) F(v) given as
@ v f ðv i Þ f ðv  0Þ A
i¼1 i
follows:

k
!1=k
k
vk1  cv 1 XN
f ðvÞ ¼ e ; v > 0; c > 0; k > 0 (1) c¼ vk f ðvi Þ (6)
c c f ðv  0Þ i¼1 i
and
where vi is the wind speed central to bin i, N the number of bins.

k f(vi) is the frequency for wind speed ranging within bin i, and
v

FðvÞ ¼ 1  e
c
; v > 0; c > 0; k > 0; (2) f(v  0) is the probability of wind speed 0.; As in MLM, Eq. (5)
must first be solved iteratively to estimate the parameter k, then
where c is the scale parameter (unit of speed) and k is the shape using the estimated k, the parameter c is obtained by solving Eq. (6).
parameter (dimensionless).
In the literature, several methods have been used for estimating
the parameters of Weibull distribution. In this section, from these 2.3. The moment method-MOM
common methods, six different methods are revisited, and one
method developed is introduced for the first time. MOM is another technique commonly-used in Weibull param-
Also, in all the methods used in this paper, in order to develop eter estimation [21]. Weibull parameters are obtained through this
the estimation results, the zero wind speeds (calm spells) are method by equating population moments to corresponding sample
handled [2,36,37]. For this, the zero wind speeds (calm spells) are moments as in the following [21,31]:
firstly removed from the measurements; the fitting method is then
v ¼ cGð1 þ 1=kÞ (7)
applied to the remaining data. As the last step, the zero wind speeds
are supposed to be re-introduced into the results to properly h i1=2
determine other parameters as well as the energy yield. On the S ¼ c Gð1 þ 2=kÞ  G2 ð1 þ 1=kÞ (8)
other hand, considering the related studies of Tuller and Brett [36]
and Carta et al. [2], which put forth no favorable advantage of this
where v and S are the sample mean and standard deviation of wind
step; it is projected that this would not influence the final results so
speed calculated respectively, as follows:
the last step is skipped in this study but the first two steps are
applied. 1X n
v¼ v (9)
n i¼1 i
2.1. The maximum likelihood method-MLM
" #1=2
1 X n
Although MLM has been a common method for estimating the S¼ ðv  vÞ2 (10)
parameters of Weibull distribution, it needs extensive iterative n  1 i¼1 i
calculation [23]. This method is performed in two stages. In the first
stage, the shape parameter k is estimated using the following and also G() is the gamma function defined by:
equation iteratively [23,29]:
Z∞
0 11 GðaÞ ¼ xa1 ex dx (11)
BPn vk lnðv Þ Pn lnðv ÞC 0
B i i C
k ¼ B i¼1
Pn i k  i¼1 C (3)
@ i¼1 v
n A To estimate the parameters k and c, using MOM, like MLM and
i
MMLM, the solutions of Eqs. (7) and (8) are obtained using nu-
merical iterations.
where vi is ith observed wind speed data and n is the number of all The estimation methods considered so far require extensive
observed non-zero wind speed data. After obtaining the value of iterative computations, however, the following methods, which
304 I. Usta / Energy 106 (2016) 301e314

Table 1
Details of the regions for which the wind speed data were analyzed in this study.

Stations Latitude (N) Longitude (E) Altitude (m) Climate Wind speed measurement

Cide 41 520 32 560 36 Black sea climate with warm, wet summers Recording by using a cup generator anemometer.
and cool to cold, wet winters. The average The hourly data over the average of 12 data with
annual temperature is 13.4  C. continuous 5 min. periods corresponding to the 1 h.
original data.
Keban 38 470 38 440 808 Continental (Mediterranean) climate with Recording by using a cup generator anemometer.
warm to hot, dry summers and mild to cool, The hourly data over the average of 12 data with
wet winters. The average annual temperature is 14.1  C. continuous 5 min. periods corresponding to the 1 h.
original data.

will now be presented, use no iterative calculations to estimate where v is the sample mean of wind speed, and v3 is the sample
parameters. mean of wind speed cubes. Then the two parameters are estimated
using the following equations [28,29]:
2.4. The graphical method-GM
3:69
k ¼ 1 þ
2 (15)
GM based on the concept of least squares is implemented to fit a Epf
straight line to wind data, where time-series data must be ranked
in ascending order. Taking the double natural logarithmic trans-
formation of the Weibull cdf, given in Eq. (2), the following equa- v
c¼ (16)
tion can be developed [22]: Gð1 þ 1=kÞ

ln½  lnð1  FðvÞÞ ¼ k lnðvÞ  k lnðcÞ (12)

This equation is considered as the simple linear model:
y ¼ b0 þ b1x where y ¼ ln[ln(1F(v))], x ¼ ln(v), b0 ¼ kln(c) and 3. Innovative estimation method for Weibull parameters
b1 ¼ k. The slope b1 and the intercept b0 estimates can be obtained
through the ordinary least square method or by manual drawing. 3.1. The probability weighted moments based on power density
Finally, the Weibull parameters are estimated as: method- PWMBP

k ¼ b1 and c ¼ eb0 =b1 (13) The PWMs (probability weighted moments) were introduced by
Greenwood et al. [38]. PWMs have almost never been applied in
wind energy studies, but they have been widely-used in many other
applications.
2.5. The energy pattern factor method-PDM
The PWMs of a random variable X with a cdf, F(x), are defined as
follows:
In order to estimate the shape k and scale c parameters with this
PDM, first, the energy pattern factor Epf is calculated as [28]: h i
PWMi;r;s ¼ E X i ðFðXÞÞr ð1  FðXÞÞs (17)
v3
Epf ¼ (14) where i, r, and s are real numbers.
ðvÞ3
Depending on the form of the considered distribution, PWMs
are generally resolved in the following form [40]:

Table 2 Table 3
Descriptive statistics of monthly, seasonal and yearly wind speed data (m/s) at Cide Descriptive statistics of monthly, seasonal and yearly wind speed data (m/s) at
for 2006. Keban for 2007.

Mean Standard Variation of Skewness Kurtosis n Mean Standard Variation of Skewness Kurtosis n
(m/s) deviation coefficient (%) (m/s) deviation coefficient (%)
(m/s) (m/s)

January 3.7472 2.8525 76.1235 1.1008 3.8687 744 January 2.0164 1.1567 57.3656 0.9823 4.1882 742
February 2.9004 2.3572 81.2704 1.4490 5.2646 672 February 1.4534 0.9338 64.2526 1.0904 4.2061 671
March 3.3056 2.4474 74.0358 1.5460 6.2754 744 March 2.0181 1.5828 78.4263 1.6056 6.5722 744
April 2.0871 1.6186 77.5554 2.0095 10.5098 720 April 1.9374 1.3164 67.9504 1.0089 3.9617 720
May 2.3726 1.8640 78.5629 1.4623 5.4386 744 May 1.9015 1.3316 70.0274 1.1044 4.6293 744
June 2.4117 1.7040 70.6579 1.0309 3.8486 720 June 2.3301 1.4954 64.1747 1.2867 4.5497 720
July 3.5187 3.1861 90.5485 1.3902 4.8947 744 July 2.2003 1.5457 70.2518 1.5786 5.8895 738
August 2.4727 2.0885 84.4620 1.5120 5.8422 743 August 1.8123 1.1338 62.5649 0.9586 3.5162 733
September 3.1096 2.3425 75.3319 1.0389 3.7890 720 September 1.7730 1.0580 59.6705 0.7505 3.1180 719
October 2.2948 2.0692 90.1717 1.6815 6.1246 744 October 1.6582 1.1270 67.9662 1.3273 4.9935 744
November 3.5361 2.7671 78.2540 0.7182 2.6008 720 November 1.6239 1.0291 63.3748 1.3362 4.6749 720
December 2.9726 2.4896 83.7531 1.0651 3.8141 744 December 1.8566 1.1664 62.8264 1.2056 5.2450 744
Jan.eMar. 3.3317 2.5896 77.7262 1.3576 4.9907 2160 Jan.eMar. 1.8419 1.2893 70.0013 1.5668 7.1469 2157
Apr.eJun. 2.2913 1.7386 75.8756 1.4860 6.2007 2184 Apr.eJun. 2.0546 1.3958 67.9356 1.1756 4.6054 2184
Jul.eSep. 3.0331 2.6195 86.3648 1.5034 5.7835 2207 Jul.eSep. 1.9301 1.2802 66.3254 1.4548 6.2153 2190
Oct.eDec. 2.9279 2.5058 85.5832 1.1173 3.7070 2208 Oct.eDec. 1.7139 1.1141 65.0059 1.2942 5.0741 2208
Yearly 2.8953 2.4208 83.6122 1.4540 5.4851 8759 Yearly 1.8848 1.2793 67.8725 1.4012 5.8884 8739
 I. Usta / Energy 106 (2016) 301e314 305

Table 4 Table 5
Estimates of parameters and performance criteria based on different methods for Estimates of parameters and performance criteria based on different methods for
monthly wind speed data at Cide. monthly wind speed data at Cide.

Period Methods Parameters Criteria Period Methods Parameters Criteria

c k R2 RMSE KS PDE c k R2 RMSE KS PDE

January MLM 4.0989 1.3500 0.94831 0.01415 0.0553 0.7222 July MLM 3.6468 1.0980 0.96639 0.01234 0.0562 7.4722
MMLM 4.1168 1.3348 0.95025 0.01386 0.0534 4.7008 MMLM 3.7130 1.1234 0.95742 0.01389 0.0652 5.5334
MOM 4.0732 1.3263 0.95205 0.01364 0.0488 2.8959 MOM 3.6530 1.1059 0.96410 0.01275 0.0584 5.5794
GM 4.0348 1.4340 0.93147 0.01671 0.0673 15.4623 GM 3.5938 1.1397 0.95479 0.01438 0.0634 8.4188
PDM 4.0895 1.3557 0.94756 0.01428 0.0558 0.9087 PDM 3.7269 1.1833 0.93371 0.01732 0.0790 8.4723
PWMM 4.0595 1.3036 0.95410 0.01334 0.0433 6.0802 PWMM 3.6140 1.0714 0.97310 0.01116 0.0482 13.3705
PWMBP 3.9804 1.3036 0.95446 0.01344 0.0379 0.0001 PWMBP 3.4659 1.0714 0.97402 0.01109 0.0408 0.0001
February MLM 3.1598 1.3056 0.95739 0.01707 0.0708 7.8069 August MLM 2.6442 1.2176 0.96949 0.01609 0.0557 3.4692
MMLM 3.1648 1.2938 0.95779 0.01702 0.0693 5.3356 MMLM 2.7062 1.2445 0.96395 0.01744 0.0668 2.3258
MOM 3.1069 1.2375 0.95714 0.01708 0.0525 0.1619 MOM 2.6221 1.1885 0.97293 0.01519 0.0474 0.5840
GM 3.0885 1.4661 0.93406 0.02146 0.0893 32.8471 GM 2.6054 1.2881 0.95721 0.01932 0.0675 20.0417
PDM 3.1207 1.2618 0.95880 0.01676 0.0584 3.4320 PDM 2.6424 1.2254 0.96837 0.01639 0.0573 5.2840
PWMM 3.1142 1.2502 0.95820 0.01686 0.0556 1.7558 PWMM 2.6224 1.1891 0.97288 0.01520 0.0475 0.4837
PWMBP 3.1327 1.2502 0.95806 0.01685 0.0578 0.0001 PWMBP 2.6182 1.1891 0.97287 0.01522 0.0471 0.0001
March MLM 3.6556 1.4322 0.99343 0.00567 0.0334 9.5269 September MLM 3.4004 1.3527 0.97739 0.01128 0.0595 2.6462
MMLM 3.7217 1.3899 0.99136 0.00677 0.0403 1.5654 MMLM 3.4514 1.3474 0.97771 0.01091 0.0629 8.2902
MOM 3.6125 1.3663 0.98873 0.00742 0.0420 3.6249 MOM 3.3871 1.3412 0.97980 0.01073 0.0560 3.4137
GM 3.5983 1.5713 0.98677 0.00864 0.0461 27.2884 GM 3.3572 1.4125 0.96375 0.01480 0.0672 9.9591
PDM 3.6027 1.3455 0.98557 0.00840 0.0449 1.0774 PDM 3.4012 1.3735 0.97276 0.01242 0.0636 0.6612
PWMM 3.6328 1.4148 0.99298 0.00586 0.0358 8.9713 PWMM 3.3719 1.3097 0.98472 0.00939 0.0483 7.7931
PWMBP 3.7484 1.4148 0.99276 0.00613 0.0411 0.0001 PWMBP 3.2886 1.3097 0.98427 0.01029 0.0407 0.0001
April MLM 2.3077 1.4078 0.98363 0.01320 0.0625 18.0961 October MLM 2.4405 1.1825 0.98843 0.01100 0.0713 10.7443
MMLM 2.4027 1.3482 0.97624 0.01646 0.0740 1.4862 MMLM 2.4858 1.1951 0.98476 0.01282 0.0802 8.4361
MOM 2.2599 1.3003 0.96874 0.01817 0.0471 8.1692 MOM 2.3857 1.1106 0.99365 0.00806 0.0509 0.5578
GM 2.2593 1.6271 0.98169 0.01454 0.0847 40.9368 GM 2.3822 1.3264 0.96061 0.02032 0.0883 38.1489
PDM 2.2465 1.2642 0.95980 0.02060 0.0540 3.3318 PDM 2.4228 1.1695 0.99048 0.00994 0.0664 9.8641
PWMM 2.2854 1.3827 0.98183 0.01387 0.0546 17.3762 PWMM 2.3964 1.1267 0.99354 0.00815 0.0552 2.5184
PWMBP 2.4355 1.3827 0.97963 0.01522 0.0619 0.0001 PWMBP 2.4169 1.1267 0.99341 0.00825 0.0532 0.0001
May MLM 2.6008 1.3502 0.99659 0.00565 0.0596 8.2405 November MLM 3.7867 1.2358 0.83523 0.02728 0.0868 18.6522
MMLM 2.6254 1.3348 0.99686 0.00557 0.0605 3.1255 MMLM 3.8026 1.2222 0.84339 0.02651 0.0848 23.7355
MOM 2.5618 1.2826 0.99803 0.00429 0.0421 1.0912 MOM 3.8214 1.2880 0.79859 0.03027 0.0998 9.8617
GM 2.5500 1.4990 0.97664 0.01571 0.0770 30.0990 GM 3.7461 1.2435 0.83165 0.02775 0.0862 13.0321
PDM 2.5655 1.2917 0.99840 0.00386 0.0443 2.3470 PDM 3.8638 1.3654 0.73945 0.03478 0.1177 0.7679
PWMM 2.5709 1.3050 0.99865 0.00355 0.0474 4.1317 PWMM 3.7675 1.2104 0.85161 0.02588 0.0802 23.5190
PWMBP 2.6073 1.3050 0.99825 0.00422 0.0526 0.0001 PWMBP 3.5113 1.2104 0.85774 0.02601 0.0854 0.0001
June MLM 2.6678 1.4566 0.99459 0.00715 0.0456 0.1305 December MLM 3.1570 1.1881 0.86869 0.02926 0.0927 9.8572
MMLM 2.6990 1.4627 0.99380 0.00749 0.0506 2.8449 MMLM 3.1718 1.1681 0.87962 0.02796 0.0892 16.9261
MOM 2.6563 1.4367 0.99530 0.00667 0.0405 1.5683 MOM 3.1595 1.1991 0.86207 0.03001 0.0954 7.3291
GM 2.6361 1.5236 0.98942 0.01079 0.0536 11.1836 GM 3.1066 1.2435 0.83713 0.03292 0.1015 7.4461
PDM 2.6602 1.4516 0.99487 0.00701 0.0438 0.0516 PDM 3.2005 1.2658 0.81742 0.03465 0.1125 3.2659
PWMM 2.6520 1.4209 0.99536 0.00659 0.0371 3.3561 PWMM 3.1208 1.1466 0.89288 0.02643 0.0807 17.6051
PWMBP 2.6230 1.4209 0.99498 0.00713 0.0334 0.0001 PWMBP 2.9565 1.1466 0.89718 0.02649 0.0698 0.0001

 
PWM1;0;s ¼ E Xð1  FðXÞÞs (18) Equating the calculated sample PWMs and the population
PWMs, and making the necessary arrangements, the shape
where s is a positive integer value. It is also shown that there is a parameter k of Weibull distribution is easily estimated as the
relationship between PWMs and L-moments, which are more following equation:
descriptive statistics and are more robust to outliers [39,40].
Therefore, they produce equivalent parameter estimates, so that, in
this study, the findings based on PWMs were also used. These lnð2Þ
findings could show a parallelism with the study of Arslan et al. [33] k¼
 (20)
who used L-moments to estimate Weibull parameters. However, it
ln C
should be emphasized that the present paper differs from their
study by adapting power density, as the neglected key issue, for and C is simply calculated by:
PWMs to reach more accurate estimation of Weibull parameters for
wind speed application.
The PWMM (probability weighted moment method) is analogous v
C¼ 2
Pn (21)
to the conventional MOM. In other words, the estimates of Weibull nðn1Þ i¼1 vðiÞ ðn  iÞ
parameters based on PWMM can be obtained by equalizing the
sample PWMs with the population PWMs. In this case, the popula- where v is the sample mean of wind speed and v(i) is ith ascending
tion PWMs of Weibull distribution can be expressed as [39,40]: ordered wind speed data and n is total of all observed wind speed
data.
cGð1 þ 1=kÞ On the other hand, it is well known that the mean power density
PWM1;0;s ¼ (19)
ð1 þ sÞ1þ1=k (per unit of rotor swept area A) based on Weibull distribution (PWD)
and the mean power density of the time series wind data (PREF) are
where s is a positive integer value and G() is a gamma function. obtained respectively, as follows [7,28,35]:
306 I. Usta / Energy 106 (2016) 301e314

Cide-Jan. Cide-May.
0.25 0.35
MLE MLE
MMLE MMLE
MOM 0.3 MOM
0.2 GE GE
PDE
Probability density function

PDE

Probability density function

0.25
PWMM PWMM
PWMMBP PWMMBP
0.15
0.2

0.1 0.15

0.1
0.05

0.05

0
0 5 10 15 0
0 2 4 6 8 10 12
Wind speed (m/s)
Wind speed (m/s)

Cide-Jul. Cide-Nov.
0.25 0.25
MLE MLE
MMLE MMLE
MOM MOM
0.2 GE 0.2 GE
PDE PDE
Probability density function

Probability density function

PWMM PWMM
PWMMBP PWMMBP
0.15 0.15

0.1 0.1

0.05 0.05

0 0
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14
Wind speed (m/s) Wind speed (m/s)

Fig. 1. Histogram and pdf graphs of Weibull distribution according to the seven methods for monthly wind speed data measured at Cide.

11=3 0
v3
  c¼@ A (24)
1 3 3 Gð1 þ 3=kÞ
PWD ¼ rc G 1 þ (22)
2 k
where v3 is the sample mean of wind speed cubes.
In this study, estimating the Weibull parameters, which are used
in Eq. (20) and Eq. (24), is called the PWMBP (probability weighted
moments based on power density method). It should be emphasized
1 Xn
v3i that PWMBP is proposed and presented for the first time to estimate
PREF ¼ r (23)
2 i¼1 n Weibull parameter more accurately in wind energy applications by
this study. Through this innovative method, the parameters k and c
where r is air density influenced with pressure (altitude), tem- could be estimated easily with less computation in a simpler
perature and humidity. On the other hand, the related studies formulation expressed in analytical form. Such analytical form does
[41e43] indicated that its variation has not got significant effect on not require extensive or iterative calculations. It is also easy to apply
wind resource calculation. Thus, the air density is accepted as when observed wind is provided. Moreover, it is a more appropriate
constant and r ¼ 1.225 kg/m3 under standard conditions (sea level, method to estimate power density. These salient features show the
15  C), which is used as the typical value in literature. superiority of this method over other common methods.
When Eq. (22) and Eq. (23) are taken into account and essential Additionally, Eq (7) can be used to estimate the scale parameter c.
calculations are performed, the scale parameter c of Weibull dis- If Eq. (20) and Eq. (7) are used to estimate parameters, this method is
tribution is easily obtained as the following equitation: called the PWMM (probability weighted moments method).
 I. Usta / Energy 106 (2016) 301e314 307

Table 6 Table 7
Estimates of parameters and performance criteria based on different methods for Estimates of parameters and performance criteria based on different methods for
monthly wind speed data at Keban. monthly wind speed data at Keban.

Period Methods Parameters Criteria Period Methods Parameters Criteria

c k R2 RMSE KS PDE c k R2 RMSE KS PDE

January MLM 2.2751 1.8351 0.99824 0.00527 0.0387 2.8824 July MLM 2.4603 1.5382 0.96954 0.02031 0.0650 11.6532
MMLM 2.2899 1.7511 0.99503 0.00984 0.0425 5.3075 MMLM 2.4887 1.4977 0.96331 0.02271 0.0696 4.0391
MOM 2.2677 1.8042 0.99771 0.00614 0.0352 1.7212 MOM 2.4256 1.4456 0.95339 0.02486 0.0538 4.9421
GM 2.2559 1.9535 0.99646 0.00912 0.0491 12.0451 GM 2.4128 1.7752 0.97369 0.01881 0.0655 33.1961
PDM 2.2664 1.7829 0.99683 0.00734 0.0352 0.3542 PDM 2.4150 1.4033 0.94063 0.02788 0.0510 0.3487
PWMM 2.2696 1.8367 0.99839 0.00503 0.0372 3.6984 PWMM 2.4437 1.5344 0.96979 0.02009 0.0611 13.0483
PWMBP 2.2982 1.8367 0.99749 0.00647 0.0453 0.0001 PWMBP 2.5603 1.5344 0.96335 0.02256 0.0679 0.0001
February MLM 1.6302 1.6371 0.99819 0.00718 0.0548 3.6900 August MLM 2.0393 1.6885 0.99310 0.01129 0.0794 2.6671
MMLM 1.6566 1.5949 0.99876 0.00570 0.0567 5.3891 MMLM 2.0515 1.6152 0.99722 0.00675 0.0714 6.1036
MOM 1.6203 1.5929 0.99966 0.00344 0.0453 1.1729 MOM 2.0257 1.6399 0.99691 0.00757 0.0693 0.2873
GM 1.6103 1.7627 0.98878 0.01944 0.0682 16.6637 GM 2.0094 1.8407 0.97503 0.02416 0.0937 17.1885
PDM 1.6189 1.5786 0.99966 0.00327 0.0429 0.0541 PDM 2.0254 1.6369 0.99703 0.00739 0.0688 0.0444
PWMM 1.6223 1.6138 0.99884 0.00543 0.0491 2.9093 PWMM 2.0265 1.6481 0.99651 0.00808 0.0707 0.9369
PWMBP 1.6383 1.6138 0.99931 0.00481 0.0532 0.0001 PWMBP 2.0329 1.6481 0.99693 0.00743 0.0681 0.0001
March MLM 2.2154 1.3597 0.97781 0.01681 0.0640 10.7512 September MLM 1.9960 1.7517 0.97387 0.02177 0.0503 0.0152
MMLM 2.2275 1.3248 0.98269 0.01487 0.0586 3.7894 MMLM 2.0080 1.6801 0.97940 0.01862 0.0418 8.1061
MOM 2.1799 1.2850 0.98739 0.01267 0.0441 3.0282 MOM 1.9892 1.7275 0.97665 0.02046 0.0454 0.9385
GM 2.1732 1.5144 0.94384 0.02777 0.0872 32.1264 GM 1.9768 1.8259 0.96549 0.02673 0.0582 8.0123
PDM 2.1790 1.2823 0.98756 0.01259 0.0434 2.6536 PDM 1.9904 1.7448 0.97494 0.02137 0.0483 0.2773
PWMM 2.1924 1.3221 0.98391 0.01432 0.0533 7.8303 PWMM 1.9880 1.7107 0.97803 0.01969 0.0426 2.1659
PWMBP 2.2528 1.3221 0.98306 0.01475 0.0574 0.0001 PWMBP 1.9738 1.7107 0.97840 0.01961 0.0422 0.0001
April MLM 2.1610 1.5378 0.96514 0.02252 0.0744 1.6722 October MLM 1.8600 1.5813 0.99474 0.01014 0.0998 8.1176
MMLM 2.1759 1.4885 0.97471 0.01953 0.0675 6.4890 MMLM 1.8801 1.5230 0.99614 0.00888 0.0970 1.2876
MOM 2.1459 1.4987 0.97332 0.01972 0.0656 0.8468 MOM 1.8366 1.4983 0.99685 0.00779 0.0828 2.7451
GM 2.1300 1.6486 0.93803 0.03174 0.0890 16.0709 GM 1.8243 1.8107 0.96759 0.02872 0.1193 28.9917
PDM 2.1472 1.5065 0.97188 0.02027 0.0672 0.0773 PDM 1.8323 1.4707 0.99544 0.00937 0.0880 0.0135
PWMM 2.1442 1.4892 0.97497 0.01928 0.0637 1.8061 PWMM 1.8433 1.5471 0.99700 0.00769 0.0909 7.1668
PWMBP 2.1314 1.4892 0.97503 0.01918 0.0621 0.0001 PWMBP 1.8896 1.5471 0.99484 0.01016 0.1006 0.0001
May MLM 2.1146 1.4940 0.94446 0.02845 0.0894 2.6918 November MLM 1.8330 1.7035 0.96049 0.03062 0.0984 8.1992
MMLM 2.1241 1.4503 0.95503 0.02549 0.0836 4.3876 MMLM 1.8452 1.5964 0.95208 0.03438 0.0879 3.5287
MOM 2.0972 1.4507 0.95532 0.02529 0.0788 0.4299 MOM 1.8129 1.6170 0.95597 0.03249 0.0820 3.8262
GM 2.0821 1.6123 0.91054 0.03766 0.1027 18.6017 GM 1.7977 1.9971 0.95580 0.03392 0.1235 28.9585
PDM 2.0981 1.4550 0.95433 0.02578 0.0797 0.0379 PDM 1.8080 1.5708 0.94983 0.03473 0.0847 0.0634
PWMM 2.0962 1.4456 0.95644 0.02516 0.0778 0.9793 PWMM 1.8202 1.7027 0.96189 0.03000 0.0948 10.0461
PWMBP 2.0894 1.4456 0.95649 0.02511 0.0765 0.0001 PWMBP 1.8856 1.7027 0.95325 0.03378 0.1026 0.0001
June MLM 2.6234 1.6690 0.95354 0.02465 0.0551 7.2225 December MLM 2.0880 1.6843 0.98813 0.01403 0.0619 5.3725
MMLM 2.6401 1.6275 0.94814 0.02660 0.0558 1.6900 MMLM 2.1295 1.6456 0.98824 0.01391 0.0647 3.9809
MOM 2.5982 1.5950 0.94424 0.02711 0.0500 3.1881 MOM 2.0744 1.6325 0.98535 0.01547 0.0513 2.6750
GM 2.5768 1.9043 0.95766 0.02324 0.0686 26.2334 GM 2.0599 1.8442 0.98229 0.01963 0.0784 19.6857
PDM 2.5921 1.5573 0.93555 0.02913 0.0522 0.0763 PDM 2.0707 1.5995 0.98236 0.01693 0.0556 0.0240
PWMM 2.6073 1.6622 0.95396 0.02444 0.0510 8.3490 PWMM 2.0782 1.6706 0.98742 0.01445 0.0577 5.5565
PWMBP 2.6842 1.6622 0.94817 0.02648 0.0567 0.0001 PWMBP 2.1182 1.6706 0.98867 0.01357 0.0562 0.0001

4. Performance selection criteria

where N is the total number of bins, yj jth the observed probability
of wind speed data, xj jth the predicted probability calculated from
To determine the performance of the seven previously
Weibull distribution, y the average of yj values, Fn(v) the observed
mentioned methods in estimating the parameters of Weibull dis-
cdf of the actual data, and F(v) the cdf of Weibull distribution. Also,
tribution, the R2, RMSE, KS statistic and PDE are used. The formulae
PREF is the mean power density of the time series data and PWD is the
of the mentioned performance selection criteria are as follows:
calculated wind power based on Weibull distribution. The best
PN
2 method for estimating Weibull parameters for the actual data can
yj  xjj¼1 be determined according to the highest value of R2 and the lowest
2
R ¼1 (25)
PN
2 values of RMSE, KS and PDE [6e9].
j¼1 yj  y

0 10:5
N

X 2 5. Results and discussion
1
RMSE ¼ @ y  xj A (26)
N j¼1 j
In this section, the various computations and analyses, based on
monthly, seasonal and yearly wind speed data, are conducted to
compare the performance of the seven different methods that are
KS ¼ maxjFn ðvÞ  FðvÞj (27)
v presented in Sections 2 and 3. Firstly, details of the actual wind
speed data are provided in the following. In pursuit of the details
  concerning actual data, the results of analyses, which are obtained
P  PWD 
PDEð%Þ ¼  REF   100 (28) according to the performance selection criteria in Section 4, are
PREF
given for monthly, seasonal and yearly wind speed data. Moreover,
the performance of these different methods is compared in terms of
308 I. Usta / Energy 106 (2016) 301e314

Keban-Jan. Keban-Mar.
0.4 0.35
MLE MLE
0.35 MMLE MMLE
MOM 0.3 MOM
GE GE
0.3 PDE PDE
Probability density function

Probability density function

0.25
PWMM PWMM
0.25 PWMBP PWMBP
0.2
0.2
0.15
0.15

0.1
0.1

0.05 0.05

0 0
0 1 2 3 4 5 6 7 8 0 2 4 6 8 10
Wind speed (m/s) Wind speed (m/s)

Keban-Jul. Keban-Dec.
0.35 0.4
MLE MLE
MMLE 0.35 MMLE
0.3 MOM MOM
GE GE
PDE 0.3 PDE
Probability density function

Probability density function

0.25
PWMM PWMM
PWMBP 0.25 PWMBP
0.2
0.2
0.15
0.15

0.1
0.1

0.05 0.05

0 0
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8
Wind speed (m/s) Wind speed (m/s)

Fig. 2. Histogram and pdf graphs of Weibull distribution according to the seven methods for monthly wind speed data measured at Keban.

actual wind energy output from a wind turbine which is calculated higher mean and standard deviation values than Keban for most of
by using the probability distribution and wind power curve. the considered time periods. Regarding the coefficients of skewness
and kurtosis, Keban has the highest values compared to Cide for all
the considered time periods, except February, April, May, August,
5.1. Wind speed data
October. Therefore, it can be concluded that Cide and Keban show
different statistical characteristics of wind speed. Moreover, it can
The wind speed data (m/s) considered in this study is taken from
be deduced from Table 2 that the zero wind speeds (calm spells) are
the Turkish State Meteorological Service. The monthly, seasonal
not detected for Cide for the all considered time periods, except
and yearly wind speed data is measured and stored hourly at 10 m
August which has one zero wind. On the contrary, for Keban, the
above ground level in two different regions of Turkey; namely for
zero wind speeds are observed for five time periods. In particular,
Cide in 2006 and for Keban in 2007. The first region, Cide, a district
the number of zero wind speeds are 2 for January, 1 for February, 6
of Kastamonu, is located in the western part of Turkey's Black Sea
for July, 11 for August and lastly 1 for September as illustrated in

province, is
region. The second region Keban, a district of the Elazıg
Table 3.
located in Turkey's Eastern Anatolia region. It is known that these
regions have different climatic conditions and geographical struc-
tures. Geographical and climatic information about these regions as 5.2. Comparisons based on monthly wind speed data
well as wind speed measurement of these regions are given in
Table 1. Firstly, the performance of the PWMBP relative to the six other
The descriptive statistics of the monthly, seasonal and yearly considered methods is evaluated on the basis of monthly wind
wind speed data of Cide and Keban are given, respectively, in speed data taken from the Cide and Keban stations in Turkey.
Tables 2 and 3. The variation for different time periods and regions The estimates of Weibull parameters, and the calculated criteria
can clearly be seen in Tables 2 and 3. In other words, Cide exhibits according to these seven methods, are given in Tables 4 and 5 for
 I. Usta / Energy 106 (2016) 301e314 309

Table 8 Table 9
Estimates of parameters and performance criteria based on different methods for Estimates of parameters and performance criteria based on different methods for
seasonal and yearly wind speed data at Cide. seasonal and yearly wind speed data at Keban.

Period Methods Parameters Criteria Period Methods Parameters Criteria

c k R2 RMSE KS PDE c k R2 RMSE KS PDE

Jan.eMar. MLM 3.6487 1.3493 0.98020 0.00974 0.0471 5.4915 Jan.eMar. MLM 2.0531 1.5198 0.99875 0.00436 0.0446 10.0372
MMLM 3.6781 1.3277 0.98070 0.00966 0.0457 0.4295 MMLM 2.0700 1.4764 0.99885 0.00470 0.0414 2.7148
MOM 3.6058 1.2973 0.98134 0.00945 0.0331 0.0615 MOM 2.0316 1.4512 0.99810 0.00543 0.0296 4.9334
GM 3.5792 1.4801 0.96371 0.01360 0.0645 26.2743 GM 2.0204 1.6885 0.98570 0.01609 0.0668 27.8180
PDM 3.6130 1.3103 0.98166 0.00937 0.0363 1.8238 PDM 2.0227 1.4082 0.99471 0.00902 0.0379 0.3078
PWMM 3.6109 1.3063 0.98160 0.00939 0.0353 1.2968 PWMM 2.0419 1.5098 0.99914 0.00361 0.0407 10.4418
PWMBP 3.6266 1.3063 0.98155 0.00940 0.0365 0.0001 PWMBP 2.1183 1.5098 0.99684 0.00776 0.0564 0.0001
Apr.eJun. MLM 2.5255 1.3964 0.99738 0.00485 0.0510 8.6280 Apr.eJun. MLM 2.2949 1.5506 0.99134 0.01076 0.0589 4.5762
MMLM 2.5677 1.3789 0.99660 0.00577 0.0545 1.3491 MMLM 2.3121 1.5035 0.99484 0.00801 0.0540 3.1086
MOM 2.4923 1.3310 0.99531 0.00648 0.0348 2.4011 MOM 2.2758 1.4991 0.99511 0.00809 0.0471 1.1358
GM 2.4775 1.5499 0.98739 0.01139 0.0679 29.3581 GM 2.2584 1.6967 0.97060 0.02168 0.0757 21.2205
PDM 2.4898 1.3239 0.99469 0.00690 0.0333 1.4994 PDM 2.2736 1.4871 0.99547 0.00774 0.0447 0.0511
PWMM 2.5017 1.3591 0.99705 0.00514 0.0410 5.8367 PWMM 2.2789 1.5165 0.99426 0.00882 0.0506 2.8003
PWMBP 2.5524 1.3591 0.99666 0.00568 0.0488 0.0001 PWMBP 2.3006 1.5165 0.99553 0.00766 0.0453 0.0001
Jul.eSep. MLM 3.2205 1.1854 0.98829 0.00795 0.0479 2.0997 Jul.eSep. MLM 2.1645 1.6077 0.99333 0.01019 0.0476 8.8038
MMLM 3.2803 1.2027 0.98452 0.00912 0.0551 0.6293 MMLM 2.1817 1.5531 0.99104 0.01228 0.0495 1.0569
MOM 3.1960 1.1613 0.99141 0.00684 0.0410 1.4943 MOM 2.1443 1.5387 0.99008 0.01250 0.0392 4.5099
GM 3.1667 1.2602 0.97462 0.01193 0.0606 20.9619 GM 2.1290 1.8079 0.98403 0.01720 0.0641 26.8112
PDM 3.2304 1.2087 0.98430 0.00921 0.0535 6.3847 PDM 2.1364 1.4903 0.98412 0.01580 0.0351 0.0567
PWMM 3.1897 1.1534 0.99213 0.00665 0.0389 2.9395 PWMM 2.1533 1.6044 0.99359 0.00994 0.0445 9.9138
PWMBP 3.1591 1.1534 0.99214 0.00656 0.0370 0.0001 PWMBP 2.2296 1.6044 0.99030 0.01237 0.0545 0.0001
Oct.eDec. MLM 3.1012 1.1742 0.93756 0.02013 0.0811 7.7246 Oct.eDec. MLM 1.9279 1.6466 0.99591 0.00874 0.0867 7.3052
MMLM 3.1231 1.1695 0.93875 0.01992 0.0819 11.3069 MMLM 1.9523 1.5798 0.99449 0.01061 0.0833 2.8959
MOM 3.0933 1.1723 0.93875 0.01994 0.0800 7.4141 MOM 1.9084 1.5727 0.99460 0.01006 0.0718 3.1501
GM 3.0396 1.2543 0.90105 0.02569 0.0936 15.1349 GM 1.8922 1.8804 0.98030 0.02254 0.1082 26.8845
PDM 3.1396 1.2431 0.90067 0.02544 0.0996 4.2978 PDM 1.9038 1.5366 0.99167 0.01252 0.0656 0.0869
PWMM 3.0561 1.1247 0.95877 0.01638 0.0658 17.1815 PWMM 1.9142 1.6254 0.99639 0.00821 0.0704 7.3915
PWMBP 2.8987 1.1247 0.96234 0.01632 0.0578 0.0001 PWMBP 1.9639 1.6254 0.99465 0.01001 0.0725 0.0001
Yearly MLM 3.1162 1.2445 0.99340 0.00616 0.0492 4.7457 Yearly MLM 2.1092 1.5687 0.99812 0.00527 0.0574 8.1873
MMLM 3.1501 1.2422 0.99295 0.00639 0.0521 1.1310 MMLM 2.1289 1.5161 0.99802 0.00579 0.0531 0.2284
MOM 3.0787 1.2012 0.99752 0.00379 0.0364 0.9706 MOM 2.0880 1.5006 0.99771 0.00583 0.0425 3.6501
GM 3.0527 1.3589 0.97344 0.01273 0.0651 27.4545 GM 2.0722 1.7619 0.98290 0.01788 0.0781 27.0469
PDM 3.1010 1.2369 0.99457 0.00559 0.0460 4.6227 PDM 2.0814 1.4640 0.99518 0.00847 0.0354 0.0064
PWMM 3.0778 1.1998 0.99758 0.00375 0.0360 1.2002 PWMM 2.0957 1.5504 0.99869 0.00440 0.0518 8.1077
PWMBP 3.0656 1.1998 0.99764 0.00373 0.0348 0.0001 PWMBP 2.1556 1.5504 0.99799 0.00580 0.0537 0.0001

regard to the RMSE and KS, similar results are observed with the R2
Cide. It can be deduced from Tables 4 and 5 that, for Cide, PWMBP
for Keban. While GM exhibits worse performance than the other
apparently outperforms the other methods according to the esti-
methods, in terms of all the criteria for several months, for the
mations of power density for all the considered months. Since the
remaining months PDM and MMLM are the least efficient methods,
estimates for Weibull parameters with PWMBP are obtained based
compared to the others, according to the R2 and RMSE.
on power density, this result appeared as expected. Furthermore,
In addition, the Weibull functions corresponding to the
this result is supported by the other considered criteria. For
considered methods are plotted against the histogram of monthly
example, taking into account the R2 and RMSE criteria, it can be
wind speed data for four months of Keban in Fig. 2. It can be
seen that PWMBP or PWMM give better results than the other
deduced from Fig. 2 that the calculated Weibull function regarding
methods to estimate k and c parameters for almost all months. The
PWMBP is a good match with the actual wind speed data for all the
KS also favors PWMBP compared with the other methods for all
considered months. It should be emphasized that PWMBP provides
months except February, March, April and May. On the other hand,
accurate fitting to the empirical distribution of wind speed data
the GM gives poor estimates of Weibull parameters in terms of the
around the peak or tail, which is obviously observed in March.
PDE particularly, and all other criteria for most of the months.
To reveal the suitability of PWMBP graphically, the Weibull pdfs,
based on parameters estimated using the seven methods and the 5.3. Comparisons based on seasonal and yearly wind speed data
histogram of monthly wind speed data for four months of Cide, are
shown in Fig. 1. As seen in Fig. 1, the estimated Weibull pdf, ac- For the seasonal and yearly wind data from Cide, the estimated
cording to PWMBP, exhibits a significantly better fit to wind speed values of parameters and the results for statistical analyses, the R2,
data, and in particular to a peak of wind data, than the other RMSE, KS and PDE values are given in Table 8. It can be concluded
methods, particularly for July and November. from Table 7 that for Cide, not only does the Weibull function
Additionally, Tables 6 and 7 present the estimates of the pa- generated by PWMBP correspond to the seasonal and yearly wind
rameters and the calculated criteria corresponding to the consid- speed data more so than the other methods in terms of the R2,
ered methods for Keban. Similar to the obtained results for Cide, the RMSE and KS, but it also has the highest efficiency for all the
PDE results for Keban in Tables 6 and 7 indicate that PWMBP shows considered data according to the PDE. Moreover, it should be
the best performance, relative to other methods, for all considered emphasized that GM yields the worst performance in terms of all
months. Taking into account the R2, the Weibull distribution, esti- criteria for Cide's seasonal and yearly data.
mated by PWMBP or PWMM, provides the highest degree of fit to The results, based on the seasonal and yearly wind data for
Keban's wind speed data for most of the considered months. With Keban, are reported as estimates of parameters k, c and the
310 I. Usta / Energy 106 (2016) 301e314

Cide-Oct.-Dec. Cide-Yearly
0.35 0.25
MLE MLE
MMLE MMLE
0.3 MOM MOM
GE 0.2 GE
PDE
Probability density function

PDE

Probability density function

0.25
PWMM PWMM
PWMMBP PWMMBP
0.2 0.15

0.15
0.1

0.1

0.05
0.05

0
0 2 4 6 8 10 12 14 0
0 2 4 6 8 10 12 14 16 18
Wind speed (m/s)
Wind speed (m/s)

Fig. 3. Histogram and pdf graphs of Weibull distribution according to the seven methods for seasonal and yearly wind speed data measured at Cide.

calculated criteria values according to the seven methods shown in

Table 9. The PDE values show the superiority of PWMBP over the Zvco
T
other methods for all seasonal and yearly wind data. While PWMBP E¼ PðvÞf ðvÞdv (31)
A
exhibits good performance for the period July to September ac- vci
cording to the R2, RMSE and KS, PWMBP provides comparable
If f(v) is taken as Weibull distribution in Eq. (31), actual wind
performance with the other methods for the remaining periods.
energy output based on Weibull distribution (EWD) is calculated by
Furthermore, Figs. 3 and 4 present how the estimated Weibull
using a basic numerical integration technique.
function matches the observed wind speed histogram based on
In order to compare the performance of the seven different
seasonal and yearly data for Cide and Keban. Figs. 3 and 4 illustrate
methods in estimating the energy production from a wind tribune,
that the Weibull pdf, according to PWMBP, exhibit better fitting to
the EPE (energy production error) is given in the following equation
the wind speed data, with an obvious peak or tail, than the other
[37,44,46]:
methods for all of the considered data.
 
E  EWD 
EPEð%Þ ¼  REF   100 (32)
EREF
5.4. Performance of actual wind energy output
where EWD is the calculated energy production based on Weibull
The energy production from a wind tribune can be calculated by distribution each parameter estimation method and EREF, the
using of its power curve, which gives the relation between the wind reference energy output, is the calculated energy output using the
speed and the output power. This relation can be illustrated as the time-series wind data, which is obtained as follows [44,46];
following [37,44,45]:
1X Nv
8 EREF ¼ Pðvi Þf vi (33)
< qðvÞ vci  v  vr A i¼1
PðvÞ ¼ P vr < v  vco (29)
: r
0 otherwise where Nv is the number of observation corresponding to power
curve values, fvi is the relative frequency according to each wind
where P(v) is the output power, Pr is the rated output power, q(v) is speed vi.
the non-linear relationship between power output and wind speed. Also, in this study, the wind energy calculations are conducted
vci is the cut-in wind speed, vr is the rated wind speed vco is the cut- for two different wind turbines, since Cide and Keban show
out wind speed. different statistical characteristics of wind speed. Hence, the actual
In literature, the various models have been used for q(v) to wind energy output and EPE for Cide wind data are obtained by
describe the non-linear relation [44e48]. However, in this study, to using the first wind tribune in which has 660 KW rate power with
represent q(v); the third degree polynomial curve is used consid- 1735 m2 swept area, the cut-in, rated, and cut-out wind speed
ering its easy computability as well as its common use [46e48]: values of 4 m/s, 15 m/s, and 25 m/s, respectively. The coefficients
values of a0, a1, a2 and a3 in Eq. (30) are 0.39209, 0.24764, 0.04446
qðvÞ ¼ a0 þ a1 v þ a2 v2 þ a3 v3 (30) and 0.00169, respectively [47]. The second wind tribune used for
Keban has Pr ¼ 50 W power with A ¼ 0.65 m2, vci ¼ 1.45 m/s and
where a0, a1, a2 and a3 are coefficients calculated from the rated vr ¼ 13 m/s. The coefficients values for Eq. (30) are a0 ¼ 0.964,
power, cut-in and cut-out wind speed. a1 ¼ 2.299, a2 ¼ 1.214 and a3 ¼ 0.0572 [46].
Thus, actual energy output of a wind tribune for a preferred Tables 10 and 11 provide the EWD and EPE as calculated from Eq.
duration in per unit of rotor swept area (A) can be obtained by using (31) and Eq. (32) corresponding to the considered estimation
the power curve and pdf of wind speed f(v) [37,44,47]: methods for Cide and Keban, respectively. The EPE results for Cide
 I. Usta / Energy 106 (2016) 301e314 311

Keban-Apr.-Jun. Keban-Yearly
0.35 0.4
MLE MLE
MMLE 0.35 MMLE
0.3 MOM MOM
GE GE
PDE 0.3 PDE
Probability density function

Probability density function

0.25
PWMM PWMM
PWMBP 0.25 PWMBP
0.2
0.2
0.15
0.15

0.1
0.1

0.05 0.05

0 0
0 1 2 3 4 5 6 7 8 9 0 2 4 6 8 10
Wind speed (m/s) Wind speed (m/s)

Fig. 4. Histogram and pdf graphs of Weibull distribution according to the seven methods for seasonal and yearly wind speed data measured at Keban.

Table 10
Actual wind energy outputs and errors based on different estimation methods for monthly, seasonal and yearly wind speed data at Cide.

Period Methods EWDx103 EPE Period EWDx103 EPE Period EWDx103 EPE

January MLM 44,9305 5,4493 July 46,6326 5,5476 Jan.-Mar. 93,5158 5,7066
MMLM 46,2579 2,6559 46,8522 5,1028 98,8205 0,3577
MOM 45,4002 4,4608 46,2764 6,2690 97,5196 1,6695
GM 39,0060 17,916 42,3195 14,2835 72,7109 26,6845
PDM 44,3577 6,6546 43,4099 12,0749 96,1947 3,0054
PWMM 46,2139 2,7485 47,5209 3,7484 96,5952 2,6015
PWMBP 43,8689 7,6832 43,5504 11,7904 97,8322 1,3542
February MLM 19,8635 10,7726 August 14,6051 6,1083 Apr.-Jun. 22,3102 14,2406
MMLM 20,4210 8,2684 14,7909 4,9144 25,1110 3,4747
MOM 21,4927 3,4545 15,3129 1,5582 25,2308 3,0141
GM 13,6843 38,5297 11,5068 26,0266 13,4882 48,1522
PDM 20,7723 6,6905 14,2799 8,1994 25,6242 1,5019
PWMM 21,1130 5,1598 15,2958 1,6680 23,7092 8,8630
PWMBP 21,5087 3,3824 15,2124 2,2044 25,8046 0,8085
March MLM 28,7771 2,0974 September 24,8725 2,4983 Jul.-Sep. 86,9141 3,8950
MMLM 32,3070 9,9117 26,3170 3,1639 88,7102 1,9090
MOM 30,4741 3,6760 25,0357 1,8588 89,0486 1,5348
GM 22,4471 23,6326 21,5821 15,3971 71,6684 20,7529
PDM 31,1716 6,0490 24,0541 5,7066 83,9308 7,1938
PWMM 28,9091 1,6484 26,0282 2,0320 89,9146 0,5773
PWMBP 31,9303 8,6302 24,0489 5,7269 87,5390 3,2041
April MLM 4,7030 25,3439 October 12,1430 17,3731 Oct.-Dec. 79,7302 5,3535
MMLM 6,7737 7,5259 12,5027 14,9257 82,1398 2,4930
MOM 6,0114 4,5757 13,9397 5,1473 79,4550 5,6801
GM 2,1271 66,2340 7,1947 51,0438 63,6777 24,4092
PDM 6,5970 4,7203 12,2995 16,3082 72,1640 14,3352
PWMM 4,8566 22,9071 13,4760 8,3022 84,6241 0,4560
PWMBP 6,5066 3,2855 13,955 5,0432 72,8924 13,4704
May MLM 9,7262 16,0940 November 40,6475 3,3495 Yearly 278,8687 7,0024
MMLM 10,5142 9,2966 41,8665 0,4509 289,7270 3,3814
MOM 10,9575 5,4719 38,7992 7,7443 293,1176 2,2507
GM 6,0621 47,7040 39,0888 7,0558 207,2392 30,8896
PDM 10,7537 7,2307 36,1467 14,0513 278,8337 7,0141
PWMM 10,4600 9,7640 41,5559 1,1895 293,6782 2,0638
PWMBP 11,0519 4,6580 34,5276 17,9012 290,1780 3,2310
June MLM 8,0275 1,1813 December 27,5283 0,1480
MMLM 8,3262 2,4952 29,0053 5,5214
MOM 8,2698 1,8011 27,0070 1,7485
GM 6,4605 20,4707 23,5040 14,4923
PDM 8,0244 1,2198 24,7158 10,0837
PWMM 8,5378 5,1010 28,9076 5,1660
PWMBP 8,1416 0,2236 24,7742 9,8712
312 I. Usta / Energy 106 (2016) 301e314

Table 11
Actual wind energy outputs and errors based on different estimation methods for monthly, seasonal and yearly wind speed data at Keban.

Period Methods EWDx103 EPE Period EWDx103 EPE Period EWDx103 EPE

January MLM 1.3952 0.4653 July 2.0765 1.4748 Jan.eMar. 3.7963 2.1939
MMLM 1.4925 6.4714 2.2033 4.5399 4.0464 4.2501
MOM 1.4065 0.3373 2.1540 2.2010 3.9405 1.5228
GM 1.2796 8.7162 1.6947 19.5906 3.1429 19.0282
PDM 1.4217 1.4234 2.2099 4.8502 4.0734 4.9454
PWMM 1.3841 1.2604 2.0466 2.8943 3.7734 2.7842
PWMBP 1.4355 2.4083 2.3027 9.2531 4.1784 7.6503
February MLM 0.5173 4.1079 August 1.0982 3.3189 Apr.eJun. 5.0767 2.3428
MMLM 0.5712 5.8921 1.1844 4.2682 5.3769 3.4333
MOM 0.5321 1.3627 1.1184 1.5415 5.1776 0.4004
GM 0.4364 19.1070 0.9431 16.9734 4.3733 15.8730
PDM 0.5392 0.0397 1.1207 1.3431 5.2170 0.3558
PWMM 0.5219 3.2583 1.1124 2.0744 5.1216 1.4781
PWMBP 0.5393 0.0173 1.1227 1.1626 5.2511 1.0125
March MLM 1.8751 3.3569 September 0.9631 1.7443 Jul.eSep. 4.1490 1.7709
MMLM 1.9718 1.6247 1.0352 5.6170 4.4308 4.9018
MOM 1.9545 0.7346 0.9702 1.0153 4.2770 1.2613
GM 1.5391 20.6748 0.8872 9.4868 3.4382 18.5991
PDM 1.9585 0.9404 0.9596 2.0944 4.4192 4.6278
PWMM 1.9013 2.0048 0.9808 0.0619 4.0994 2.9432
PWMBP 2.0330 4.7821 0.9596 2.1011 4.5166 6.9340
April MLM 1.4375 2.9203 October 0.9261 6.2528 Oct.eDec. 2.8886 4.6590
MMLM 1.5291 3.2624 1.0134 2.5851 3.1817 5.0141
MOM 1.4593 1.4515 0.9703 1.7837 2.9905 1.2958
GM 1.2628 14.7163 0.7147 27.6488 2.2783 24.8035
PDM 1.4516 1.9716 1.0100 2.2399 3.0904 2.0014
PWMM 1.4688 0.8089 0.9322 5.6390 2.8796 4.9548
PWMBP 1.4452 2.4000 1.0040 1.6298 3.0685 1.2786
May MLM 1.4550 3.3579 November 0.7663 6.3991 Yearly 15.8884 2.7248
MMLM 1.5350 1.9541 0.8620 5.2941 17.0648 4.4781
MOM 1.4829 1.5083 0.8002 2.2504 16.4070 0.4507
GM 1.2576 16.4719 0.5764 29.5856 13.0308 20.2197
PDM 1.4782 1.8214 0.8304 1.4294 16.8425 3.1168
PWMM 1.4884 1.1426 0.7497 8.4189 15.8507 2.9552
PWMBP 1.4754 2.0041 0.8285 1.2032 17.0826 4.5869
June MLM 2.1939 0.8214 December 1.1992 1.8815
MMLM 2.2844 3.2688 1.3071 6.9426
MOM 2.2384 1.1902 1.2243 0.1684
GM 1.8606 15.8893 1.0323 15.5414
PDM 2.2807 3.0989 1.2510 2.3552
PWMM 2.1681 1.9902 1.1950 2.2323
PWMBP 2.3350 5.5549 1.2445 1.7381

in Table 10 indicate that out of 16 considered time periods, PWMBP methods according to the estimations of energy output for almost
shows better performance for 6 time periods, namely February, all the considered time periods. Consequently, PWMBP provides
April, May, June, October and AprileJune season than the other more accurate and efficient estimation than other methods in
methods. Following PWMBP, PWMM provide good results just for 4 estimating the parameters of Weibull distribution. In addition,
time periods. As seen the EWD results, the most actual energy PWMBP is expressed in analytical form and offers a simpler
output is yield by using PWMM in Jully with 47520.9 kWh and the calculation. Therefore, PWMBP can be applied as an improved
least energy production obtained by GM is 2127.1 kWh in April. method to estimate the Weibull parameters for many new wind
Moreover, it can be observed from Table 11 that for Keban, PWMBP energy applications.
outperforms the other methods according to the estimations of
energy output for most of the considered time periods, February,
August, October, November and OctobereDecember season. The 6. Conclusions
monthly energy output values for Keban based on the second
tribune with 50 W power rate range from 436.4 kWh to 2333 kWh. The following conclusions can be drawn from the presented
On the other hand, the worst performance is provided by GM with study:
regard to all energy output for both Cide and Keban's wind speed
data. 1. The PWMBP is developed and proposed for the first time to
It can also be concluded from the results of analysis based on the estimate Weibull parameters for wind energy applications in
actual wind speed data that PWMBP shows a good performance for this study.
most of the considered time periods which exhibit different sta- 2. The superiority of PWMBP is summarized over the other
tistical characteristics of wind speed, since it appears among the methods: (i) having a simple formulation expressed in analytical
best ones in the considered methods in terms of the performance form; (ii) not requiring extensive numerical calculations or
criteria. The Weibull distribution function regarding PWMBP pro- iterative procedures; (iii) being easy and practical in obtaining
vides generally a good fitting to the empirical distribution of real the estimations of Weibull parameters in the case of the
wind speed data. Moreover, PWMBP outperforms the other observed wind speed; and (iv) being more appropriate for the
 I. Usta / Energy 106 (2016) 301e314 313

estimation of power density and energy output using power [11] Islam MR, Saidur R, Rahim NA. Assessment of wind energy potentiality at
Kudat and Labuan, Malaysia using Weibull distribution function. Energy
curve.
2011;36(2):985e92.
3. The performance of PWMBP is compared with six most common [12] Ouammi A, Dagdougui H, Sacile R, Mimet A. Monthly and seasonal assessment
methods, MLM, MMLM, MOM, GM, PDM and PWMM for actual of wind energy characteristics at four monitored locations in Liguria region
wind data, based on different time periods and regions. (Italy). Renew Sust Energy Rev 2010;14(7):1959e68.
[13] Safari B, Gasore J. A statistical investigation of wind characteristics and wind
4. The suitability and accuracy of the proposed PWMBP are eval- energy potential based on the Weibull and Rayleigh models in Rwanda.
uated in terms of different goodness of fit criteria commonly Renew Energy 2010;35(12):2874e80.
used in wind energy literature, as well as in terms of graphical [14] Keyhani A, Ghasemi-Varnamkhasti M, Khanali M, Abbaszadeh R. An assess-
ment of wind energy potential as a power generation source in the capital of
illustration for monthly, seasonal and yearly wind speed data Iran, Tehran. Energy 2010;35:188e201.
measured in various regions of Turkey, Cide and Keban. More- [15] Arslan O. Technoeconomic analysis of electricity generation from wind energy
over, the performance of these different methods is compared in in Kutahya, Turkey. Energy 2010;35:120e31.
[16] Ucar A, Balo F. Evaluation of wind energy potential and electricity generation
terms of wind energy output from a wind turbine which is at six locations in Turkey. Appl Energy 2009;26:1864e71.
calculated by using the probability distribution and wind power [17] Akdag S, Guler O. Calculation of wind energy potential and economic analysis
curve. by using Weibull distribution-a case study from Turkey. Part 1: determination
of Weibull Parameters. Energy Sources Part B 2009;4(1):1e8.
5. It can be observed that PWMBP apparently outperforms the
[18] Akpinar EK, Akpinar S. Determination of the wind energy potential for Maden,
other methods according to estimations capability of power Turkey. Energy Convers Manag 2004;45(18e19):2901e14.
density for all the considered data. Furthermore, the Weibull [19] Gebizlioglu OM, Şenog
lu B, Kantar YM. Comparison of certain value-at-risk
estimation methods for the two-parameter Weibull loss distribution.
function generated by the PWMBP has the highest efficiency for
J Comput Appl Math 2011;235(11):3304e14.
most of the considered data with regard to the rest of the [20] Kantar YM, Senoglu B. A comparative study for the location and scale pa-
considered criteria, R2, RMSE and KS. For instance, The R2 results rameters of the Weibull distribution with given shape parameter. Comput
for Cide indicate that out of 12 considered months, PWMBP Geosci 2008;34:1900e9.
[21] Justus CG, Mikhail A. Height variation of wind speed and wind distribution
shows better performance for 4 months than the other methods. statistics. Geophys Res Lett 1976;3:261e4.
Following PWMBP, PWMM provides good results just for 3 [22] Justus CG, Hargraves WR, Mikhail A, Graber D. Methods for estimating wind
months. speed frequency distributions. J Appl Meteorol 1978;17:350e3.
[23] Stevens MJ, Smulders PT. The estimation of the parameters of the Weibull
6. It can be concluded that while PWMBP yields the best perfor- wind speed distribution for wind energy utilization purposes. Wind Eng
mance in terms of energy production estimation for 5 time 1979;3:132e45.
periods of Keban's wind speed data, PWMBP exhibits better [24] Seguro JV, Lambert TW. Modern estimation of the parameters of the Weibull
wind speed distribution for wind energy analysis. J Wind Eng Ind Aerod
performance for 6 time periods than the other methods out of 2000;85:75e84.
16 considered time periods for Cide according to the EPE. On the [25] Dorvlo ASS. Estimating wind speed distribution. Energy Convers Manag
other hand, PWMM outperform just for 4 time periods for the 2002;43:2311e8.
[26] Silva GR. Características de vento da Regia ~o Nordeste, analise, modelagem e
aforementioned conditions. ~es para projetos de centrais eolicas [Master's thesis, Recife Pe]. 2003.
aplicaço
7. Consequently, PWMBP provides more accurate and efficient [27] Jowder FAL. Wind power analysis and site matching of wind turbine gener-
estimation of Weibull parameters for wind energy applications ators in Kingdom of Bahrain. Appl Energy 2009;86:538e45.
[28] Akdag S, Dinler A. A new method to estimate Weibull parameters for wind
and, therefore, it can be applied as an improved method to es-
energy applications. Energy Convers Manag 2009;50:1761e6.
timate the Weibull parameters for many new wind energy [29] Chang TP. Performance comparison of six numerical methods in estimating
applications. Weibull parameters for wind energy application. Appl Energy 2011;88(1):
272e82.
[30] Costa Rocha PA, de Sousa RC, de Andrade CF, da Silva MEV. Comparison of
Acknowledgments seven numerical methods for determining Weibull parameters for wind en-
ergy generation in the northeast region of Brazil. Appl Energy 2012;89:
395e400.
The authors would like to thank the anonymous referees and [31] Azad AK, Rasul GM, Yusaf T. Statistical diagnosis of the best Weibull methods
Associate Editor for their helpful suggestions and valuable for wind power assessment for agricultural applications. Energies 2014;7:
3056e85.
comments. [32] de Andrade CF, Neto HFM, Costa Rocha PA, da Silva MEV. An efficiency
comparison of numerical methods for determining Weibull parameters for
wind energy applications: a new approach applied to the northeast region of
References Brazil. Energy Convers Manag 2014;86:801e8.
[33] Arslan T, Bulut YM, Yavuz AA. Comparative study of numerical methods for
[1] Celik AN. A statistical analysis of wind power density based on the Weibull determining Weibull parameters for wind energy potential. Renew Sust En-
and Rayleigh models at the southern region of Turkey. Renew Energy ergy Rev 2014;40:820e5.
2004;29(4):593e604. [34] Akdag S, Dinler A. A novel energy pattern factor method for wind speed
[2] Carta JA, Ramirez P, Velazquez S. A review of wind speed probability distri- distribution parameter estimation. Energy Convers Manag 2015;106:
butions used in wind energy analysis e case studies in the Canary Islands. 1124e33.
Renew Sust Energy Rev 2009;13:933e55. [35] Celik AN. A techno-economic analysis of wind energy in southern Turkey. Int J
[3] Morgan EC, Matthew L, Vogel RM, Baise LG. Probability distributions for Green Energy 2007;4:233e47.
offshore wind speeds. Energy Convers Manag 2011;52(1):15e26. [36] Tuller SE, Brett AC. The characteristics of wind velocity that favour the fitting
[4] Chang TP. Estimation of wind energy potential using different probability of a Weibull distribution in wind speed analysis. J Clim Appl Meteorol
density functions. Appl Energy 2011;88(5):1848e56. 1984;23:124e34.
[5] Soukissian T. Use of multi-parameter distributions for offshore wind speed [37] Carrillo C, Cidra s J, Díaz-Dorado E, Obando-Montan ~ o AF. An approach to
modeling: the Johnson SB distribution. Appl Energy 2013;111:982e1000. determine the Weibull parameters for wind energy analysis: the case of
[6] Usta I, Kantar YM. Analysis of some flexible families of distributions for esti- Galicia (Spain). Energies 2014;7:2676e700.
mation of wind speed distributions. Appl Energy 2012;89(1):355e67. [38] Greenwood JA, Landwehr JM, Matalas NC, Wallis JR. Probability weighted
[7] Kantar YM, Usta I. Analysis of the upper-truncated Weibull distribution for moments: definition and relation to parameters of several distributions
wind speed. Energy Convers Manag 2015;96:81e8. expressible in inverse form. Water Resour Res 1979;15:1049e54.
[8] Shamilov A, Kantar YM, Usta I. Use of MinMaxEnt distributions defined on [39] Hosking JRM. L-moments: analysis and estimation of distributions using
basis of MaxEnt method in wind power study. Energy Convers Manag linear combinations of order statistics. J Roy Stat Soc Ser B 1990;52:105e24.
2008;49:660e77. [40] Usta I. Different estimation methods for the parameters of the extended burr
[9] Kantar YM, Usta I. Analysis of wind speed distributions: wind distribution XII distribution. J Appl Stat 2013;40(2):397e414.
function derived from minimum cross entropy principles as better alternative [41] Ramirez P, Carta JA. Influence of the data sampling interval in the estimation
to Weibull function. Energy Convers Manag 2008;49:962e73. of the parameters of the Weibull wind speed probability density distribution:
[10] Fazelpour F, Soltani N, Soltani S, Rosen MA. Assessment of wind energy po- a case study. Energy Convers Manag 2005;46:2419e38.
tential and economics in the north-western Iranian cities of Tabriz and [42] Tar K. Some statistical characteristics of monthly average wind speed at
Ardabil. Renew Sust Energy Rev 2015;45:87e99. various heights. Renew Sust Energy Rev 2008;12:1712e24.
314 I. Usta / Energy 106 (2016) 301e314

[43] Safari B. Modeling wind speed and wind power distributions in Rwanda. [46] Celik AN. Energy output estimation for small-scale wind power generators
Renew Sust Energy Rev 2011;15:925e35. using Weibull-representative wind data. J Wind Eng Ind Aerod 2003;91:
~ o AF, Cidr
[44] Carrillo C, Obando Montan as J, Díaz-Dorado E. Review of power 693e707.
curve modelling for wind turbines. Renew Sust Energy Rev 2013;21: [47] Chang T-J, Wu Y-T, Hsu H-Y, Chu C-R, Liao C-M. Assessment of wind char-
572e81. acteristics and wind turbine characteristics in Taiwan. Renew Energy
[45] Lydia M, Kumar SS, Selvakumar AI, Prem Kumar GE. Review of power 2003;28(6):851e71.
curve modelling for wind turbines. Renew Sust Energy Rev 2013;21: [48] Akdag SA, Guler OA. Comparison of wind turbine power curve models. Energy
572e81. Sources Part A 2011;33:2257e63.