Electrical Power and Energy Systems 103 (2018) 497–510

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

A probabilistic assessment approach for wind turbine-site matching T

a,b,⁎ b,⁎ c a
N. Aghbalou , A. Charki , S.R. Elazzouzi , K. Reklaoui
a
Engineering, Innovation Management and Industrial Systems Research Laboratory, FSTT-University Abdelmalek ESSAADI, Tangier, Morocco
b
Systems Engineering Research Laboratory, ISTIA-University of Angers, Angers, France
c
Laboratory of Information & Communication Technology, ENSAT-University Abdelmalek ESSAADI, Tangier, Morocco

A R T I C LE I N FO A B S T R A C T

Keywords: This article provides a new methodology for wind turbine-site matching by using a probabilistic approach. The
Wind turbine performance random behavior of the wind speed climate and the uncertainties of wind turbine characteristics are important to
Probabilistic performance assessment take into account in models used to evaluate the performance of the wind turbine. The proposed formulation of
Random wind conditions the wind turbine-site matching is derived based on the probabilistic reliability assessment approach. It was
Wind turbine-site matching
experimented using different power curve approximation models, for different random conditions, using time
series of wind speed in two sites in Morocco: Dakhla and Essaouira. A comparison based on methods used in
literature for the estimation of two-parameter of the Weibull function to fit the wind speed distribution is also
carried out. The results revealed that the introduced performance indicators are less sensitive to the models used
to approximate the wind power curves compared to the deterministic conventional indicator that leads to dif-
ferent rankings and problems of over-sizing or under-sizing. However, those performance indicators are more
sensitive to the variation of the wind speed distribution parameter’s and can help on accurately estimate the
wind power. Moreover, the proposed formulation allows a global sensitivity analysis using Sobol’s indices to
observe the influence of each input parameter on the observed variances of the performance of a wind turbine. A
numerical application illustrates the interpretation of sensitivity indices and shows the impact of the wind speed
and the rated wind speed on the variance of the wind turbine performance. This method can help wind energy
developers and manufacturers to optimally select WTGs for their future project and accurately forecast the
performance of their WTGs for monitoring and maintenance scheduling under uncertainty.

1. Introduction Performance Index (NTPI) [5], Amount Of Annual Energy Production

Due to the effects of fossil fuel energy on the environment, devel-
opers have been obliged to seek out other cleaner energy arising from
the sustainable resources. Wind energy is perceived as the most af-
fordable, mainly due to its generated electricity capacity and compe-
titive cost [1]. However, the intermittent and the stochastic nature of
the wind speed remain the most challenging aspect against the massive
integration of this technology. In fact, the characteristics of the wind
speed vary with time and space, and the optimal selection of a wind
turbine should be conducted carefully in order to find the best Wind
Turbine Generator (WTG) characteristics with respect to the wind speed
availability. It is a critical step from the reliability and cost-benefit point
of view. In literature, many studies have been reported on the optimal
site matching of wind turbines. Most of these studies are based on
Capacity Factor (CF)[1], Cost of Energy (COE), Annual Net Profit (ANP)
[2], the Wind Turbine-Site Matching Index (TSMI)[3], Turbine cost
Index and Integrated Matching Index (TIIMI) [4], Normalized Turbine
(AEP) and Wind Turbine Performance Index (TPI). Lee et al. [2] as-
sessed the effects of local wind speed and wind turbine power char-
acteristics on the optimum hub height, they concluded that the op-
timum hub height decreased as the mean wind speed and wind shear
exponent increased. Moreover, it was shown that rated power and cut-
in speed had little effect, where as rated speed and cut-out speed had
much greater effects on optimum hub height. Chang et al. [5] demon-
strated that the wind turbine performance would be better if the wind
speed Weibull distribution concentrates at higher speed range or
especially if it is greater than the wind turbine’s rated speed. WTG in-
stalled at high height could produce more than at lower height because
of high rated wind speed occurrence. This may decrease the net annual
profit by increasing the initial capital cost of installation and the op-
eration and maintenance cost which may results due to failure caused
by excessive wind speed [2]. Chang et al. [5], highlighted that if the
wind turbine is selected with low rated speed, it would operate with a
high capacity, in this case too much energy would be lost in high wind

⁎
Corresponding authors at: Engineering, Innovation and Management of Industrial Systems Research Team, FSTT-University Abdelmalek ESSAADI, Tangier, Morocco.
E-mail addresses: nihad.aghbalou@gmail.com (N. Aghbalou), abderafi.charki@univ-angers.fr (A. Charki).

https://doi.org/10.1016/j.ijepes.2018.06.018
Received 22 November 2017; Received in revised form 16 April 2018; Accepted 5 June 2018
Available online 15 June 2018
0142-0615/ © 2018 Elsevier Ltd. All rights reserved.
N. Aghbalou et al. Electrical Power and Energy Systems 103 (2018) 497–510

Nomenclature Pelec (v ) Electric power of wind turbine

pfc Probability of full charge operating
α Hellman’s wind shear component pf Probability of failure
βHL The Euclidean distance from P∗ to the origin of the stan- po Probability of operating
dard space ppc Probability of partial charge operating
χ2 Chi-square test statistic Pr Rated power
γf Indicator function Pr (.) Probability operator
μ Average value Si The first order Sobol indice measuring the variance con-
∇ The gradient operator tribution of variable x i to the variance of the response
Φ(.) The standard Gaussian cumulative distribution function STi The total order Sobol indice measuring the interaction
ϕn (U ) n-dimensional standard Gaussian probability distribution contributions of x i with all the other variables
function U Vector of standard Gaussian random variables
σ Standart deviation (u1, u2, …, un ) mutually independent and uncorrelated
c Scale parameter U∗ The coordinate vector of P∗
CF Capacity factor v Wind speed
f (v ) Weibull probability density function V (h) Wind speed at the height h
fX (X ) Joint density function of random variables X Vc Cut-in speed
g (v ) Non-linear part of power curve Vf Cut-out speed
G (X ) Performance function Vm Mean wind speed
H Wind turbine hub height Vref Measured wind speed at href
H (U ) The performance function expressed in the standard space Vr Rated speed
href Reference height above ground level X Vector of stochastic variables ( x1, x2 , …, x n )
k Shape parameter z0 Surface roughness length
P∗ The most probable failure point in the standard space
Pc Critical wind power

speed in this site. Contrariwise, if the wind turbine is selected with high
rated speed, the wind turbine may extract most of the wind energy, but
its capacity becomes low.
In the aforementioned researches, the uncertain aspect of different
variables such as WTG characteristics or wind shear component was not
considered. It was observed and mentioned in [6,7] that the expected
typical power curve is often not ideally followed. This could be derived
by errors in wind speed measurement, the lack of knowledge of the long
term variation of the characteristics of WTG, the lack of accuracy in
power loss expectation for non-nominal wind conditions as the power
curves should be carefully calibrated to the wind conditions in the
desired area according to the standard IEC 61400-12 [8].
Few scientific studies were based on a probability or reliability
theory. Jung et al. [9] proposed a Bayesian approach to account for
uncertainty in wind speed, air density, surface roughness exponent and
power performance of the turbine using the Annual Energy Production
(AEP). Melchers [10] developed an interesting method to evaluate the
system performance in using the reliability theory. He proposed the
estimation of the failure probability of a system in taking into account
the uncertainties influence of input variables (operating conditions) on
physical models. He used several methods (First or Second Order Re-
liability Methods (FORM, SORM), Monte Carlo) for the assessment of
the failure probability. Charki et al. [11,12] and Titikpina et al. [13]
used the structural reliability approach for studying the performance of
other energy systems and mechanical components. They showed that
the method proposed by Melchers [10] is suitable for assessing the
performance and the reliability of a system.
This paper proposes the adaptation of the method of structural re-
liability in wind energy. It provides a new methodology for wind tur-
bine-site matching issue taking into account the uncertainties derived
from the random behavior of wind speed climate and uncertainties
likewise concern WTG characteristics and models used to approximate
the WTG power curve. Various methods (Maximum Likelihood Method
(MLM), Least Squares Estimation Method (LSEM), two Maximum
Entropy methods (MaxEnt1) and (MaxEnt2 ), Cuckoo Search optimiza-
tion method (CS) and Particle Swarm Optimization method (PSO)) have
been selected and tested to estimate the statistical parameters of wind
speed distribution, to calculate the wind power curve in order to assess
the performance and the reliability of wind turbines from two sites in
Morocco: (Dakhla et Essaouira) operating in random conditions.
Several models used in literature have also been selected and tested.
Results show that the developed methodology based on the structural
reliability analysis and prediction [14,15], is suitable to improve the
effectiveness of a WTG. Moreover, the proposed methodology high-
lights the main variables that affect mostly the performance of wind
turbine.
2. Wind energy and turbine characteristics
2.1. Wind speed distribution modeling
The first fundamental step in assessing wind power potential con-
sists of the statistical analysis of the wind speed in the selected sites.
The main purpose is to expect the probability of certain wind speeds
occurrence. Several models may be found in the literature. Wang et al.
[16] reviewed and compared some non-parametric and parametric
(unimodal distributions and multimodal distributions) models for wind
speed probability distribution. They concluded that the non-parametric
Kernel density distribution outperforms all of the selected parametric
models in terms of the fitting, accuracy and the operational simplicity.
Ouarda et al. [17] found that among the one-component parametric
distributions, the Kappa and generalized Gamma distributions provide
the best fit to the wind speed data at all heights. However, Ouarda et al.
[17] underline that the Weibull distribution performs better than the
Generalized Extreme Value and 3-parameter Lognormal. In a later
work, Ouarda et al. [18] reviewed different criteria used in the field of
wind energy to compare the goodness-of-fit of candidate probability
density functions to wind speed records, and discussed their advantages
and disadvantages. The 2-parameter Weibull distribution remains the
most commonly accepted distribution in the specialized literature on
wind energy and other renewable energy sources [19–21]. The general
form of Weibull probability density function is given by:
k v k−1 v k
f (v ) = ( ) exp ⎡−⎛ ⎞ ⎤
c c ⎢ ⎝c⎠ ⎥ (1)
⎣ ⎦
where c is the scale parameter, k is the shape parameter and v is the

498
N. Aghbalou et al. Electrical Power and Energy Systems 103 (2018) 497–510

c=6
a)
Density of probability

k v (m/s)
Fig. 2. Power curve modeling objectives.
c=8
b) of these methods are: Graphical Method (GM); Standard Deviation
Method (STDM); Power Density Method (PDM); Modified Maximum
Likelihood Method (MMLM); Equivalent Energy Method (EEM); Method
of Moments (MOM); Maximum Likelihood Method (MLM); Least
Density of probability

Squares Estimation Method (LSEM); stochastic heuristic optimization

algorithms (GA, CS, PSO , etc) [4,16,23,24], and others [25]. For this
study, Maximum Likelihood Method (MLM), Least Squares Estimation
Method (LSEM), Maximum Entropy methods (MaxEnt1, MaxEnt 2 ) as
described in [26–28], Cuckoo Search optimization method (CS) [29]
and Particle Swarm Optimization method (PSO) [30] are used.

2.2. WTG characteristics and power curve modeling

In general, WTGs are characterized by their power curves which

k v (m/s) specify the different operational speeds: cut-in speed Vc , cut-out speed
(or furling)Vf and rated speed Vr . This helps to predict the wind energy
c=12 production for a particular WTG, to be (or) placed on a site, without the
c) technical details about components information in the wind turbine
generating system [31]. The power curve is required for many other
tasks regarding the WTG during its life cycle [31–33] see Fig. 2. It is
usually modeled by means of approximations [19] to depict relation-
Density of probability

ship between output power and hub height wind speed without any
other details about WTG dynamics. The power curve is expressed as
follows:

k v (m/s)
Fig. 1. Wind speed weibull’s distribution with different shape and scale values.
wind speed.
Fig. 1 shows the variations of the wind speed distribution with the
Weibull parameters c and k. It is evident from Fig. 1 that the scale
parameter c informs about the occurrence of highest wind speed and
the duration of the extreme wind speeds. However, the shape paramete
k informs about the regularity of the wind regime: as k becomes lower,
the wind regime becomes more turbulent [22], contrariwise, it becomes
regular and concentrated around the mean value when k becomes
larger. To estimate Weibull parameters from wind speed time series
data, it is necessary to use a statistical estimator. The accuracy of the
estimated values of c and k depends directly on the available data. In
particular, it depends on the time resolution, size and errors in the
observed sample of wind speed measurements. Several estimators have
been used in the statistical modeling of wind speed distribution. Some
⎧ g (v ) Vc < v < Vr
Pelec (v ) = Pr 1 Vr < v < Vf
⎨
⎩ 0 otherwise (2)
where Pr is the rated power, g (v ) is the non-linear function that re-
presents the wind turbine output between the cut-in and rated speed.
Table 1 summarizes selected models mostly recommended in lit-
erature to be experienced in this paper.
As illustrated in Fig. 3, below the cut-in speed Vc , the wind turbine
cannot produce because there is no sufficient torque to rotate the wind
turbine. Above the cut-out speed Vf , some devices force the shutdown of
the wind turbine to avoid the extreme loads. Between the cut-in speed
Vc and the rated speed Vr , the wind turbine starts to produce at the
reduced performance operating zone1 (partial charge) and a MPPT
device (Maximum Power Point Tracking) is used to maximize the
captured energy. Between Vr and Vf , the WTG works at its rated per-
formance operating zone2 (full charge). Carillo et al. [19] have carried
out a review of equations commonly used to represent the power curve
of variable wind speed turbine generators, by means of four mathe-
matical equations for representing the non-linear region between cut-in
and rated speed: polynomial; exponential; cubic; approximate model.
They experimented their study on nearly 200 commercial wind turbines
and concluded that exponential and cubic models are the best when the
coefficient of determination and the error in energy density are

499
N. Aghbalou et al. Electrical Power and Energy Systems 103 (2018) 497–510

Table 1
Power curve approximations and corresponding capacity factor.
g(v) approximations Capacity factor Ref.

g1 (v ) =
v − Vc e (−V f / c )k − e (−Vr / c )k [35]
Vr − Vc CF = −e (−Vf / c )k
(Vr / c )k − (Vc / c )k

v 2 − Vc2 2c 2 Vr k 2 ⎞ V k 2 [36]
g2 (v ) =
Vr2 − Vc2
CF = −e (−Vf / c )k +
k (Vr2 − Vc2)
Γ ( ) ⎧⎨⎩γ ⎛⎝ ( ) ,
2
k c k
−γ ⎛ c , ⎞ ⎫
⎠ c
⎝k ⎬( ) ⎠⎭

v k − Vck e (−V f / c )k − e (−Vr / c )k [37]

g3 (v ) = CF = −e (−Vf / c )k
Vrk − Vck (Vr / c )k − (Vc / c )k

v3 3
3Γ ⎛ ⎞
[38]
g4 (v ) = Vc 3 k 3 V k 3
Vr3 CF = −e (−Vf / c )k + ( ) e (−V /c)
Vr c
k +
V 3⎨
k⎛ r⎞ ⎩ ⎝
( )
⎝ k ⎠ ⎧ ⎛ Vr
γ
c
, ⎞−γ ⎛ c , ⎞ ⎫
k
⎠ ⎝ c k ⎬( )
⎠⎭
⎝c ⎠
n [39,1]
g5 (v ) = ∑0 ai v i ici Vr k i ⎞ V k i
CF = −e (−Vf / c )k + ∑1 ai
n
k
Γ
i
k ( ) ⎧⎨⎩γ ⎛⎝ ( ) ,
c k
−γ ⎛ c , ⎞ ⎫
⎠ c
⎝k ⎬( ) ⎠⎭

Table 4
Example of uncertainty related to the wind speed assessment.
Uncertainty source Type of Factors affecting the
uncertainty uncertainty

Wind velocity Measurement Type of anemometer,

measurement Calibration, Reliability, mast
position
Time step ahead Statistical and Limited available data, Spatial
forecasting (horizon) model extrapolation models (MCP)
Spatial extrapolation and model CFD model, roughness, height
wind shear of the mast

Table 5
Wind turbine models.
wt Model H (m) Vc (m/s) Vr (m/s) Vf (m/s) Pr (kW)

#1 Nordex 40 3 14 25 600
#2 Gamessa G52 RCC 44 4 15 25 800
Fig. 3. Power curves models at an elevation of 50 m. #3 Gamessa G52 850 44 4 16 25 850
#4 Gamessa G58 850 44 4 15 21 850
#5 Vestas-V39-45 45 4 18 25 600
Table 2 #6 Vestas-V44-45 45 5 17 20 600
#7 Vestas-V47-45 45 4 17 25 660
Wind shear exponent in various area types [52].
#8 Jeumon j48-CL1 46 4 14 25 750
Area type α #9 Vestas Mic. M1500 46 4 15 25 600
#10 Enercon E-33 50 3 12.5 25 330
Smooth hard ground,calm water 0.1 #11 Vestas-V52 55 4 17 25 850
Tall grass on level ground 0.15 #12 Vestas-V66 60 4 16 25 1750
High crops, hedges and shrubs 0.2 #13 ZEPHYROS-z72 65 3 16 25 2000
Wooded countryside, many trees 0.25 #14 GE 2–7 70 3.5 16 25 2700
Small town with trees and shrubs 0.3 #15 Suzlon s88 80 4 14 25 2100
Large city with tall buildings 0.4 #16 GE 2–5 85 3.5 15 25 2500
#17 GE 2–3 100 3 14 25 2300
#18 Nordex N100 100 3 13 20 2500
Table 3
Values of surface roughness length [53].
Table 6
Terrain description z 0 (mm) Description of the studied sites.
Very smooth, ice or mud 0.01 Site Lat. Long. Height μ (m/s) σ (m/s) Min (m/s) Max (m/s)
Calm open sea 0.20
Blown sea 0.50 Dakhla 24 −15.3 50 m 7.77 2.3 0.02 14.95
Snow surface 3.00 Essaouira 32 −9.33 50 m 5.07 2.5 0.03 17.70
Lawn grass 8.00
Rough pasture 10.00
Fallow field 30.00 considered. Thapar et al. [34] classified models for predicting the
Corps 50.00
Few trees 100.00
performance of wind turbines into two categories: models based on
Many trees, hedges, few buildings 250.00 fundamental equations of power available in the wind and models
Forest and woodlands 500.00 based on the concept of power curve of wind turbine. Conclusions
Suburbs 1500.00 drawn from their study indicate that models based on a presumed shape
Canters of cities with tall buildings 3000.00
of power curve give satisfactory response for higher annual average
wind speeds, however methods in which actual power curve of a wind

500
N. Aghbalou et al. Electrical Power and Energy Systems 103 (2018) 497–510

Fig. 4. Flowchart of the method proposed.

Table 7 basis of data used for modeling. They concluded that parametric
The estimated parameters of the Weibull model using various estimation polynomial approximation based power curve models have been used
methods at Dakhla. widely since they are simple to use and can be for designing small
Estimator systems and nonparametric methods (i.e. Artificial Neural Network
(ANN), Adaptive Neuro-Fuzzy Inference System (ANFIS)) perform well
H Parameter LSE MLE MaxEnt1 MaxEnt2 CS PSO among other models for power prediction and online monitoring ap-
plications. Fig. 3 shows a graphical comparison of the five listed models
50 m Vm 7.7608 7.8092 7.9106 7.5349 7.9883 7.9886
k 3.3298 3.8157 3.9 3.3 4.1924 4.1906 in Table 1 and the power curve of Enercon E-33 (H = 50 m, Vc = 3 m/s,
c 8.7019 8.5850 8.74 8.4 8.7896 8.7901 Vr = 12.5 m/s, Vf = 25 m/s and Pr = 330 kW) commercial wind tur-
RMSE 0.017 0.011 0.008 0.1528 0.0071 0.0070 bine. Graphically, these models don’t fit ideally the power curve, and
underestimate the expected performance of this WTG except g1 (v ) that
overestimates it. g3 (v ) is a model that depend on the shape parameter k
Table 8 of the wind speed distribution in the site.
The estimated parameters of the Weibull model using various estimation
methods at Essaouira.
2.3. Capacity factor in wind energy assessment
Estimator
As mentioned before, many studies were reported in wind energy
H Parameter LSE MLE MaxEnt1 MaxEnt2 CS PSO
assessment and optimal WTG site matching. The capacity factor (CF) is
50 m Vm 5.7371 5.6980 5.80 5.394 5.8420 5.8300 a dimensionless indicator defined as the wind turbine’s annual elec-
k 2.2823 2.4180 2.47 2.26 2.4176 2.4150 tricity yield (in kWh) divided by the electricity output if the wind
c 6.4764 6.4267 6.54 6.09 6.8592 6.5758 turbine had operated at its rated power output for the entire year (i.e.
RMSE 0.0067 0.0061 0.0055 0.011 0.011 0.0051
the installed power times 365 days times 24 h) [5,40–42]. Reasonable
capacity factors range from 0.25 to 0.30, while a very good capacity
factor would be 0.40 [43]. The CF index is expressed as:
turbine is based on the method of least squares and cubic spline in-
terpolation give accurate results for wind turbines having smooth V
∫Vc f Pelec (v ) f (v ) dv
power curve, whereas, for the turbines which do not have so smooth CF =
Pr (3)
power curve, model based on method of least squares is best suited.
Sohoni et al. [32] classified the power curve modeling methods into It was observed by [37], that CF is solely a function of hub height
discrete, deterministic/probabilistic, parametric/nonparametric, sto- and independent on the wind turbine rated power. In fact, replacing Eq.
chastic methods and Other classification could be conducted in the (2) in Eq. (3) CF can be written as:

501
N. Aghbalou et al. Electrical Power and Energy Systems 103 (2018) 497–510

Table 9 Table 11
Calculated probabilities and capacity factors for Dakhla. Statistics of indicators when changing wind power curve approximation
models.
Model Estimator pfc ppc po CF
DAKHLA ESSAOUIRA
LSE 3.66E−2 9.35E−1 9.37E−1 3.79E−1
MLE 1.61E−2 9.66E−1 9.67E−1 3.77E−1 Model Statistc po CF po CF
g1 (v ) MaxEnt1 1.88E−2 9.66E−1 9.67E−1 3.80E−1
MaxEnt2 2.54E−2 9.42E−1 9.43E−1 3.74E−1 g1 (v ) μ 9.62E−1 3.79E−01 8.43E−1 3.16E−01
CS 1.37E−2 9.75E−1 9.75E−1 3.81E−1 σ 1.73E−02 2.73E−03 1.89E−2 1.04E−02
PSO 8.90E−3 9.80E−1 9.80E−1 3.81E−1
g 2 (v ) μ 9.61E−01 3.90E−01 8.43E−1 2.12E−01
LSE 3.63E−2 9.35E−1 9.37E−1 3.93E−1
MLE 1.59E−2 9.66E−1 9.67E−1 3.82E−1 σ 1.65E−2 1.47E−02 1.86E−2 1.81E−02
g 2 (v ) MaxEnt1 1.86E−2 9.66E−1 9.67E−1 3.97E−1
g3 (v ) μ 9.60E−1 2.43E−01 8.43E−1 1.89E−01
MaxEnt2 2.52E−2 9.42E−1 9.43E−1 3.64E−1
CS 1.35E−2 9.75E−1 9.75E−1 4.02E−1 σ 1.64E−2 2.29E−02 1.86E−2 2.00E−02
PSO 1.35E−2 9.76E−1 9.76E−1 4.02E−1
g 4 (v ) μ 9.61E−1 2.97E−01 8.42E−1 1.50E−01
LSE 3.62E−2 9.34E−1 9.36E−1 2.83E−1 σ 1.64E−2 1.15E−02 2.56E−2 1.49E−02
MLE 1.58E−2 9.65E−1 9.66E−1 2.32E−1
g3 (v ) MaxEnt1 1.84E−2 9.65E−1 9.66E−1 2.40E−1 g5 (v ) μ 9.48E−1 3.07E−01 7.95E−1 1.47E−01
MaxEnt2 2.1E−2 9.41E−1 9.42E−1 2.56E−1 σ 2.06E−2 1.31E−02 2.21E−2 1.57E−02
CS 1.33E−2 9.74E−1 9.74E−1 2.24E−1
PSO 1.34E−2 9.74E−1 9.74E−1 2.24E−1

LSE 3.62E−2 9.35E−1 9.37E−1 3.06E−1 Table 12

MLE 1.58E−2 9.66E−1 9.67E−1 2.88E−1 Statistics of indicators when changing the Weibull distribution parameter’s of
g 4 (v ) MaxEnt1 1.85E−2 9.66E−1 9.67E−1 3.02E−1
wind speed estimator.
MaxEnt2 2.51E−2 9.42E−1 9.43E−1 2.78E−1
CS 1.34E−2 9.75E−1 9.75E−1 3.05E−1 po CF
PSO 1.34E−2 9.75E−1 9.75E−1 3.04E−1
Estimator μ σ μ σ
LSE 3.62E−2 9.17E−1 9.20E−1 3.14E−1
MLE 1.58E−2 9.53E−1 9.54E−1 2.97E−1
Dakhla LSE 9.34E−1 7.67E−3 3.35E−1 4.82E−2
g5 (v ) MaxEnt1 1.85E−2 9.54E−1 9.55E−1 3.12E−1
MLE 9.64E−1 5.63E−3 3.15E−1 6.38E−2
MaxEnt2 2.51E−2 9.22E−1 9.55E−1 2.84E−1 MaxEnt1 9.64E−1 5.18E−3 3.26E−1 6.35E−2
CS 1.34E−2 9.66E−1 9.24E−1 3.16E−1 MaxEnt2 9.39E−1 8.62E−3 3.11E−1 5.39E−2
PSO 1.35E−2 9.66E−1 9.66E−1 3.16E−1 CS 9.73E−1 3.89E−3 3.26E−1 7.03E−2
PSO 9.75E−1 5.02E−3 3.25E−1 7.03E−2

Table 10 Essaouira LSE 8.19E−1 2.42E−2 2.05E−1 6.78E−2

Calculated probabilities and capacity factors for Essaouira. MLE 8.37E−1 2.17E−2 1.95E−1 7.14E−2
MaxEnt1 8.48E−1 2.1 E−2 2.01E−1 7.09E−2
Model Estimator pfc ppc po CF MaxEnt2 8.00E−1 2.46E−2 1.80E−1 7.11E−2
CS 8.52E−1 1.91E−2 2.25E−1 6.55E−2
LSE 1.16E−2 8.29E−1 8.31E−1 3.16E−1 PSO 8.32E−1 2.70E−2 2.11E−1 6.92E−2
MLE 7.04E−3 8.46E−1 8.47E−1 3.14E−1
g1 (v ) MaxEnt1 7.36E−3 8.57E−1 8.85E−1 3.18E−1
MaxEnt2 6.46E−3 8.10E−1 8.11E−1 2.99E−1 Table 13
CS 1.44E−2 8.59E−1 8.61E−1 3.31E−1
Values of Weibull parameters of wind speed distribution based on PSO algo-
PSO 9.40E−3 8.50E−1 8.51E−1 3.20E−1
rithm for studied sites.
LSE 1.15E−2 8.29E−1 8.31E−1 2.13E−1
MLE 6.98E−3 8.46E−1 8.47E−1 2.05E−1 Essaouira Dakhla
g 2 (v ) MaxEnt1 7.29E−3 8.56E−1 8.57E−1 2.13E−1
H(m) c k c k
MaxEnt2 6.40E−3 8.10E−1 8.11E−1 1.84E−1
CS 1.43E−2 8.58E−1 8.60E−1 2.40E−1
40 8.51 4.16 6.48 2.4
PSO 9.33E−3 8.50E−1 8.51E−1 2.17E−1
44 8.63 4.18 6.52 2.4
LSE 1.15E−2 8.29E−1 8.31E−1 1.92E−1 45 8.65 4.18 6.53 2.4
MLE 6.97E−3 8.45E−1 8.46E−1 1.75E−1 46 8.68 4.19 6.54 2.41
g3 (v ) MaxEnt1 7.28E−3 8.56E−1 8.57E−1 1.78E−1 55 8.78 4.18 6.58 2.41
MaxEnt2 6.40E−3 8.10E−1 8.11E−1 1.66E−1 60 8.89 4.2 6.63 2.41
CS 1.43E−2 8.58E−1 8.60E−1 2.08E−1 65 9.01 4.19 6.68 2.41
PSO 9.33E−3 8.50E−1 8.51E−1 2.17E−1 70 9.11 4.17 6.73 2.41
80 9.21 4.16 6.75 2.41
LSE 1.15E−2 8.29E−1 8.40E−1 1.52E−1 85 9.38 4.17 6.82 2.4
MLE 6.96E−3 8.46E−1 8.53E−1 1.43E−1 100 9.46 4.19 6.85 2.41
g 4 (v ) MaxEnt1 7.27E−3 8.56E−1 8.63E−1 1.49E−1
MaxEnt2 6.39E−3 8.11E−1 8.18E−1 1.28E−1
CS 1.43E−2 8.58E−1 8.72E−1 1.74E−1 Vr Vf
PSO 5.35E−3 8.01E−1 8.07E−1 1.51E−1 CF = ∫V
c
g (v ) f (v ) dv + ∫Vr
f (v ) dv
(4)
LSE 1.15E−2 7.80E−1 7.83E−1 1.50E−1
MLE 6.97E−3 7.97E−1 7.98E−1 1.40E−1 Table 1 presents the capacity factors expressions corresponding to
g5 (v ) MaxEnt1 7.27E−3 8.09E−1 8.10E−1 1.46E−1
each approximating model g (v ) . According to Albadi et al. [1], the CF
MaxEnt2 6.39E−3 7.55E−1 7.57E−1 1.24E−1
CS 1.43E−2 8.15E−1 8.18E−1 1.72E−1
may underestimate the wind potential. However, using the cubic mean
PSO 5.35E−3 8.01E−1 8.07E−1 1.51E−1 wind speed in estimating Weibull probability distribution function
parameters c and k, may result in a better estimation of the CF than
using the arithmetic wind speed or the root mean wind speed.

502
N. Aghbalou et al. Electrical Power and Energy Systems 103 (2018) 497–510

Table 14
Wind turbine ranking for the site of Dakhla.
wt g1 (v ) g 2 (v )

pfc R° ppc R° po R° CF R° pfc R° ppc R° po R° CF R°

#1 169.8E−6 9 985.4E−3 3 985.4E−3 3 3.54E−1 3 155.7E−6 9 985.1E−3 3 985.1E−3 3 2.95E−1 6

#2 32.6E−6 12 956.8E−3 14 956.8E−3 14 3.12E−1 11 29.5E−6 12 957.5E−3 13 957.5E−3 13 2.40E−1 11
#3 1.6E−6 15 956.0E−3 15 956.0E−3 15 3.00E−1 13 1.3E−6 15 957.0E−3 15 957.5E−3 15 2.09E−1 14
#4 30.1E−6 13 957.5E−3 12 957.5E−3 13 2.87E−1 16 27.2E−6 13 957.2E−3 14 957.2E−3 14 2.40E−1 12
#5 1.0E−9 18 958.3E−3 11 958.3E−3 11 2.77E−1 17 64.2E−13 18 958.0E−3 12 958.0E−3 12 1.63E−1 17
#6 62.5E−9 17 897.6E−3 17 897.6E−3 17 2.32E−1 18 50.1E−9 17 898.4E−3 17 898.4E−3 17 1.60E−1 18
#7 68.9E−9 16 958.9E−3 10 958.9E−3 10 2.89E−1 15 52.2E−9 16 958.8E−3 10 958.8E−3 11 1.85E−1 16
#8 5.63E−8 6 959.9E−3 8 958.9E−3 8 3.26E−1 8 536.3E−6 6 959.3E−3 9 959.3E−3 9 2.84E−1 7
#9 51.2E−6 11 959.8E−3 9 959.8E−3 9 3.13E−1 10 46.7E−6 11 960.6E−3 8 960.6E−3 8 2.44E−1 10
#10 8.9E−3 5 980.2E−3 4 980.4E−3 4 3.81E−1 1 13.5E−3 4 975.5E−3 5 975.8E−3 5 4.02E−1 2
#11 1.6E−6 14 967.5E−3 7 967.5E−3 7 2.94E−1 14 1.4E−6 14 968.1E−3 7 968.1E−3 7 1.99E−1 15
#12 82.0E−6 10 971.3E−3 6 971.3E−3 6 3.08E−1 12 74.9E−6 10 970.5E−3 6 970.5E−3 6 2.34E−1 13
#13 2.16E−8 8 991.5E−3 1 991.5E−3 1 3.38E−1 5 197.9E−6 8 991.4E−3 1 991.4E−3 1 2.62E−1 8
#14 4.56E−8 7 985.4E−3 2 985.4E−3 2 3.26E−1 9 426.6E−6 7 985.9E−3 2 985.9E−3 2 2.59E−1 9
#15 21.7E−3 3 957.2E−3 13 958.1E−3 12 3.36E−1 6 21.4E−3 3 958.1E−3 11 959.0E−3 10 3.44E−1 4
#16 13.8E−3 4 974.6E−3 5 975.0E−3 5 3.41E−1 4 8.7E−3 5 980.0E−3 4 980.2E−3 4 3.16E−1 5
#17 54.8E−3 2 940.6E−3 16 943.9E−3 16 3.72E−1 2 54.3E−3 2 941.0E−3 16 9.44E−5 16 3.93E−1 3
#18 1.17E−5 1 878.0E−3 18 892.3E−3 18 3.35E−1 7 117.1E−3 1 878.0E−3 18 892.3E−3 18 4.56E−1 1

Table 15
Continued.
wt g3 (v ) g 4 (v )

pfc R° ppc R° po R° CF R° pfc R° ppc R° po R° CF R°

#1 149.7E−6 9 983.4E−3 3 983.4E−3 3 1.24E−1 7 151.6E−6 9 985.3E−3 2 985.3E−3 2 1.98E−1 7

#2 28.0E−6 12 957.1E−3 10 957.1E−3 11 9.58E−2 10 28.5E−6 12 958.2E−3 12 958.2E−3 13 1.58E−1 11
#3 1.2E−6 15 955.0E−3 13 956.8E−3 13 7.28E−2 14 1.3E−6 15 956.9E−3 14 956.9E−3 14 1.30E−1 14
#4 25.8E−6 13 956.8E−3 12 954.9E−3 12 9.51E−2 11 26.2E−6 13 956.4E−3 15 956.4E−3 15 1.58E−1 12
#5 51.7E−13 18 954.9E−3 14 954.9E−3 14 4.44E−2 18 55.5E−13 18 960.3E−3 8 960.3E−3 8 9.16E−2 18
#6 44.1E−9 17 895.2E−3 17 895.2E−3 17 5.39E−2 17 46.2E−9 17 899.1E−3 17 899.1E−3 17 9.82E−2 17
#7 45.5E−9 16 954.7E−3 15 954.7E−3 15 5.72E−2 16 47.6E−9 16 958.3E−3 11 958.3E−3 12 1.10E−1 16
#8 522.3E−6 6 958.8E−3 9 958.8E−3 9 1.31E−1 6 527.3E−6 6 959.3E−3 10 959.3E−3 10 1.98E−1 6
#9 44.5E−6 11 959.2E−3 8 959.2E−3 8 9.79E−2 9 45.2E−6 11 959.8E−3 9 959.8E−3 9 1.61E−1 10
#10 13.4E−3 4 974.2E−3 5 974.5E−3 5 2.24E−1 2 13.4E−3 4 975.4E−3 5 975.7E−3 5 3.04E−1 2
#11 1.2E−6 14 966.5E−3 7 966.5E−3 7 6.38E−2 15 1.3E−6 14 967.3E−3 7 967.3E−3 7 1.20E−1 15
#12 71.7E−6 10 970.9E−3 6 970.9E−3 6 8.75E−2 13 72.8E−6 10 970.8E−3 6 970.9E−3 6 1.50E−1 13
#13 190.4E−6 8 991.3E−3 1 991.3E−3 1 9.49E−2 12 192.7E−6 8 991.4E−3 1 991.6E−3 1 1.64E−1 9
#14 413.7E−6 7 984.5E−3 2 984.5E−3 2 9.88E−2 8 417.8E−6 7 984.4E−3 3 984.8E−3 3 1.66E−1 8
#15 21.3E−3 3 957.0E−3 11 957.9E−3 10 1.83E−1 4 21.4E−3 3 958.2E−3 13 959.1E−3 11 2.55E−1 4
#16 8.6E−3 5 980.2E−3 4 980.4E−3 4 1.43E−1 5 8.7E−3 5 981.1E−3 4 981.3E−3 4 2.19E−1 5
#17 54.1E−3 2 941.1E−3 16 944.3E−3 16 2.08E−1 3 54.2E−3 2 941.2E−3 16 944.4E−3 16 2.91E−1 3
#18 116.9E−3 1 878.0E−3 18 892.3E−3 18 2.78E−1 1 117.0E−3 1 878.4E−3 18 892.6E−3 18 3.59E−1 1

2.4. Factors affecting the uncertainty height h.

Additional to wake effect caused by the interaction in the wind
farm, components degradation and mechanical/electrical losses that
affect the expected performance of wind turbine, several factors influ-
ence the accuracy of wind energy assessment. More attention should
also be paid to wind climate measurement parameters particularly
average wind speed, wind shear, air density and turbulence [32,44]. In
general, wind speed is measured from one or several meteorological
masts. In order to obtain the variation of wind speed with height at
target site from a reference site, a number of method of extrapolation
and long-term correction have been proposed in the literature [45–47].
The IEC standard [8] recommended the use of the power law where the
expression is as follows:
α
h ⎞
V (h) = Vref ⎜⎛ ⎟
⎝ href ⎠
where href is a reference height above ground level (AGL) used for fit-
ting the wind profile, Vref is the measured wind speed at href , α is
Hellman’s wind shear (or friction) exponent and V (h) is wind speed at
The value of Hellman’s wind shear exponent, is not constant, varies
with the atmosphere stability, can be estimated based on Table 2 or
calculated based on surface roughness length z 0 i.g. by means of Eq. (6)
proposed by Spera and Richards [48]:
z
α = ⎜⎛ 0 ⎟⎞ [1−0.55∗log(Vref )]
h
⎝ ref ⎠ (6)
The surface roughness length z 0 may be estimated based on Table 3.
Hence, different sources of uncertainties can affect the expected per-
formance of WTG. According to [49], uncertainties can be categorized
into aleatory (physical) and epistemic (model, statistical, measurement)
uncertainties. Table 4 reminds some source of uncertainty regarding
wind speed. Jung et al. [9] analyzed the AEP estimation and identified
the effect of the uncertainties that exist due to limited available data
and the inherent uncertainty in wind speed, air density, surface
(5)
roughness exponent and power performance of the turbine. Authors in
[50] demonstrated that uncertainty of power curve measurements, for
flat or complex terrain, is of the order of 6–8% while the statistical
variation of the power curves for a given type of WTG is in the range of

503
N. Aghbalou et al. Electrical Power and Energy Systems 103 (2018) 497–510

Table 16
Continued.
wt g5 (v )

pfc R° ppc R° po R° CF R°

#1 152.6E−6 9 948.5E−3 10 948.5E−3 10 1.99E−1 7

#2 28.7E−6 12 953.3E−3 9 953.3E−3 9 1.62E−1 10
#3 1.3E−6 15 941.2E−3 11 941.2E−3 11 1.29E−1 14
#4 26.5E−6 13 954.0E−3 8 954.0E−3 8 1.62E−1 11
#5 57.5E−13 18 885.5E−3 17 885.5E−3 18 8.27E−2 18
#6 47.2E−9 17 897.1E−3 16 897.1E−3 16 1.00E−1 17
#7 48.7E−9 16 918.5E−3 15 918.5E−3 15 1.04E−1 16
#8 529.5E−6 6 959.0E−3 4 959.0E−3 4 2.09E−1 6
#9 45.6E−6 11 956.9E−3 6 956.9E−3 6 1.65E−1 8 Fig. 7. Variation of the po indicator for the selected wind turbines for the site of
#10 13.5E−3 4 966.1E−3 2 966.5E−3 2 3.16E−1 2 Dakhla (see Tables 14, 15 and 16).
#11 1.3E−6 14 936.1E−3 13 936.1E−3 13 1.16E−1 15
#12 73.3E−6 10 960.6E−3 3 960.6E−3 3 1.51E−1 13
#13 194.0E−6 8 940.8E−3 12 940.8E−3 12 1.60E−1 12
#14 419.9E−6 7 956.8E−3 7 946.8E−3 7 1.65E−1 9
#15 21.4E−3 3 957.6E−3 5 958.5E−3 5 2.69E−1 4
#16 8.7E−3 5 968.9E−3 1 969.1E−3 1 2.24E−1 5
#17 54.2E−3 2 929.1E−3 14 932.2E−3 14 3.00E−1 3
#18 117.0E−3 1 872.2E−3 18 887.1E−3 17 3.71E−1 1

Fig. 8. Variation of the CF indicator for the selected wind turbines for the site of
Dakhla (see Tables 14, 15 and 16).

optimal structure can be solved by a mathematical formulation of a

Fig. 5. Variation of the pfc indicator for the selected wind turbines for the site of
Dakhla (see Tables 14, 15 and 16).
performance function [14,15]. This performance function describes, in
general, the ability of the studied structure to perform a functional
requirement expected operational and environmental conditions for a
particular period of time. The realization of this performance function
G (X ) informs about the satisfaction of the required function. X denotes
stochastic variables that includes physical, model, statistical and mea-
surement uncertainties. In structural engineering, for instance, it may
refer to geometric variables, strength of material, loads, etc. Mathe-
matically, the probability of the dissatisfaction of the performance
function can be written as follows:

pf = Pr(G (X ) ⩽ 0) = ∫G (X )⩽0 fX (x ) dx (7)

where fX (X ) is the joint density function of random variables X and

Pr (.) is the probability operator.
Methods to calculate the integral in Eq. (7) are classified into two
methods: approximate methods and simulation methods.

3.1.1. Approximate methods

Fig. 6. Variation of the ppc indicator for the selected wind turbines for the site of
Dakhla (see Tables 14, 15 and 16).
2–3%. This is due to the fact that existed power curve models were
developed to be independently of site-specific conditions [51].
In our proposed method, the following variables: cut-in speed,
rated-speed, cut-out speed, rated power, hub height mast, wind speed
velocity, mean wind speed, surface roughness length and wind shear
component are considered to evaluate the performance function.
3. Principle of method proposed
3.1. Probabilistic performance assessment
In structural engineering, the problem of designing a reliable and
The main purpose of the structural probabilistic assessment methods
is computing the probability of failure Eq. (7). However, when the
number of the stochastic variables is important and the shape of limit
state function is complicated, computing this integral becomes difficult.
Approximate reliability methods, called also Gradient-based methods,
have been developed to approximate this integral by linearization the
limit state using Taylor series expansion at the design point. The first
order reliability method (FORM) is considered to be the most commonly
used, its accuracy depends on three parameters, i.e. the curvature ra-
dius at the design point, the number of random variables and the first-
order reliability index [54]. The main steps of the FORM method are
summarized here after:
Isoprobabilistic transformation of physical variables X to un-
correlated standard Gaussian variables U. Several transformations may
be applied. For example, Rosenblatt transformation, Nataf

504
N. Aghbalou et al. Electrical Power and Energy Systems 103 (2018) 497–510

Table 17
Wind turbine ranking for the site of Essaouira.
wt g1 (v ) g 2 (v )

pfc R° ppc R° po R° CF R° pfc R° ppc R° po R° CF R°

#1 1.42E−3 7 8.47E−1 5 8.47E−1 5 2.90E−1 4 1.39E−3 7 8.46E−1 4 8.46E−1 5 1.66E−1 4

#2 5.64E−4 10 7.27E−1 16 7.72E−1 16 2.49E−1 11 5.53E−4 10 7.27E−1 15 7.27E−1 15 1.24E−1 11
#3 1.64E−4 14 7.28E−1 15 7.28E−1 15 2.36E−1 14 1.60E−4 14 7.27E−1 16 7.27E−1 17 1.08E−1 14
#4 5.64E−4 11 7.27E−1 17 7.27E−1 17 2.38E−1 13 5.53E−4 11 7.27E−1 17 7.27E−1 16 1.24E−1 12
#5 1.11E−5 18 7.29E−1 13 7.29E−1 14 2.13E−1 17 1.06E−5 18 7.29E−1 13 7.29E−1 14 8.49E−2 17
#6 4.63E−5 17 5.84E−1 18 5.84E−1 18 1.87E−1 18 4.49E−5 17 5.83E−1 18 5.83E−1 18 7.65E−2 18
#7 4.67E−5 16 7.29E−1 14 7.29E−1 13 2.24E−1 16 4.50E−5 16 7.29E−1 14 7.29E−1 13 9.58E−2 16
#8 1.83E−3 6 7.31E−1 12 7.31E−1 12 2.64E−1 8 1.81E−3 6 7.30E−1 12 7.32E−1 12 1.45E−1 7
#9 5.97E−4 8 7.32E−1 11 7.32E−1 11 2.50E−1 10 5.86E−4 8 7.32E−1 11 7.32E−1 11 1.25E−1 10
#10 9.40E−3 2 8.50E−1 4 8.51E−1 4 3.20E−1 1 9.33E−3 2 8.50E−1 3 8.51E−1 3 2.17E−1 2
#11 7.41E−5 15 7.44E−1 10 7.44E−1 10 2.26E−1 15 7.16E−5 15 7.43E−1 10 7.43E−1 10 9.77E−2 15
#12 3.52E−4 13 7.51E−1 9 7.51E−1 9 2.40E−1 12 3.44E−4 13 7.51E−1 9 7.51E−1 9 1.14E−1 13
#13 4.62E−4 12 8.71E−1 2 8.71E−1 3 2.69E−1 7 4.50E−4 12 8.71E−1 2 8.71E−1 2 1.36E−1 8
#14 5.92E−4 9 8.24E−1 7 8.24E−1 7 2.57E−1 9 5.80E−4 9 8.24E−1 7 8.24E−1 7 1.29E−1 9
#15 5.35E−3 4 7.67E−1 8 7.68E−1 8 2.72E−1 6 5.31E−3 4 7.67E−1 8 7.68E−1 8 1.59E−1 5
#16 2.78E−3 5 8.32E−1 6 8.32E−1 6 2.74E−1 5 2.74E−3 5 8.32E−1 6 8.32E−1 6 1.53E−1 6
#17 8.12E−3 3 8.80E−1 1 8.81E−1 1 3.05E−1 2 8.05E−3 3 8.80E−1 1 8.81E−1 1 1.90E−1 3
#18 1.77E−2 1 8.70E−1 3 8.72E−1 2 3.02E−1 3 1.76E−2 1 8.46E−1 5 8.49E−1 4 2.21E−1 1

Table 18
Continued.
wt g3 (v ) g 4 (v )

pfc R° ppc R° po R° CF R° pfc R° Pp c R° po R° CF R°

#1 1.39E−3 7 8.46E−1 5 8.46E−1 5 1.38E−1 4 8.20E−4 7 8.48E−1 5 8.49E−1 5 1.06E−1 5

#2 5.52E−4 10 7.27E−1 15 7.27E−1 15 1.04E−1 10 3.00E−4 11 7.29E−1 15 7.29E−1 15 8.12E−2 9
#3 1.59E−4 14 7.27E−1 16 7.27E−1 17 8.82E−2 14 1.10E−4 14 7.27E−1 16 7.27E−1 17 6.69E−2 14
#4 5.52E−4 11 7.27E−1 17 7.27E−1 16 1.04E−1 11 2.90E−4 12 7.27E−1 17 7.27E−1 16 8.08E−2 10
#5 1.05E−5 18 7.28E−1 14 7.28E−1 14 6.62E−2 17 1.00E−5 17 7.30E−1 14 7.30E−1 14 4.72E−2 18
#6 4.46E−5 17 5.83E−1 18 5.83E−1 18 6.27E−2 18 3.00E−5 16 5.85E−1 18 5.85E−1 18 4.84E−2 17
#7 4.47E−5 16 7.29E−1 13 7.29E−1 13 7.63E−2 16 1.00E−5 18 7.33E−1 11 7.33E−1 12 5.61E−2 16
#8 1.81E−3 6 7.30E−1 12 7.30E−1 12 1.23E−1 7 1.03E−3 6 7.32E−1 13 7.32E−1 13 1.00E−1 6
#9 5.84E−4 8 7.31E−1 11 7.32E−1 11 1.04E−1 9 3.10E−4 10 7.33E−1 12 7.33E−1 11 8.17E−2 8
#10 9.32E−3 2 8.49E−1 4 8.59E−1 4 1.87E−1 2 4.75E−3 2 8.55E−1 4 8.56E−1 4 1.54E−1 2
#11 7.12E−5 15 7.43E−1 10 7.43E−1 10 7.75E−2 15 4.00E−5 15 7.42E−1 10 7.42E−1 10 5.73E−2 15
#12 3.42E−4 13 7.51E−1 9 7.51E−1 9 9.22E−2 13 2.20E−4 13 7.53E−1 9 7.53E−1 9 7.06E−2 13
#13 4.48E−4 12 8.70E−1 2 8.70E−1 3 1.07E−1 8 3.40E−4 9 8.71E−1 3 8.71E−1 3 7.82E−2 11
#14 5.77E−4 9 8.23E−1 7 8.23E−1 7 1.03E−1 12 4.00E−4 8 8.24E−1 7 8.24E−1 7 7.75E−2 12
#15 5.30E−3 4 7.67E−1 8 7.76E−1 8 1.36E−1 5 2.81E−3 4 7.70E−1 8 7.71E−1 8 1.11E−1 4
#16 2.74E−3 5 8.32E−1 6 8.32E−1 6 1.27E−1 6 1.55E−3 5 8.34E−1 6 8.34E−1 6 9.84E−2 7
#17 8.03E−3 3 8.80E−1 1 8.88E−1 1 1.59E−1 3 4.32E−3 3 8.84E−1 1 8.85E−1 1 1.26E−1 3
#18 1.76E−2 1 8.70E−1 3 8.87E−1 2 1.90E−1 1 9.50E−3 1 8.79E−1 2 8.80E−1 2 1.55E−1 1

transformation, Rackwitz-Fiesler transformation, Chen-Lind transfor- improved Hasofer-Lind-Rackwitz-Feissler (HLRF) iterative algorithm.
mation, Wu-wurshing transformation, etc. This transformation consists The so-called iHLRF algorithm aims to determine the reliability index
in rewritting the probability of failure as follows: βHL that represent the Euclidean distance from P∗ to the origin O of the
standard space (O, P∗) , and given by:
pf = ∫H (U )⩽0 ϕn (U ) du1 du2. .dun (8) T
∇H (U ∗) ⎤
βHL = −⎡ U∗
where ϕn (U ) is the n-dimensional standard Gaussian probability dis- ⎢
⎣ ‖∇H (U ∗)‖ ⎥
⎦ (11)
tribution function given by: where ∇ is the gradient operator.
1 1 As the first order Taylor expansion of the performance function at
ϕn (U ) = n exp(− U T U )
2π 2 2 (9) the design point can be expressed as:
Hfo (U ) = H (U ∗) + (U −U ∗) ∇H (U ∗) (12)
and H (U ) is the performance function expressed in the standard space.
Finding the most probable failure point i.e. design point P∗ in the Hence the approximation of the probability failure:
standard space, which represents the solution of the following optimi-
zation problem pf∗ = ∫H fo (U ) ⩽ 0
ϕn (U ∗) du = Φ(−βHL)
(13)
1
⎧U ∗ = argmin 2 U T U where Φ, denotes the standard Gaussian cumulative distribution func-
⎨ H (U ) = 0 tion.
⎩ (10)
∗
where U ∗ is the coordinate vector of P in the standard space. 3.1.2. Simulation methods
This constrained optimization problem is solved by means of Simulation methods are based on the calculation of the performance

505
N. Aghbalou et al. Electrical Power and Energy Systems 103 (2018) 497–510

Table 19
Continued.
wt g5 (v )

pfc R° ppc R° po R° CF R°

#1 5.90E−4 7 7.08E−1 11 7.08E−1 11 9.87E−2 6

#2 3.00E−4 10 7.16E−1 7 7.16E−1 7 7.72E−2 9
#3 1.00E−4 14 6.78E−1 13 6.78E−1 13 6.7E−2 14
#4 2.30E−4 12 7.15E−1 9 7.15E−1 9 7.72E−2 10
#5 1.00E−5 17 5.62E−1 18 5.62E−1 18 3.77E−2 18
#6 2.00E−5 16 5.78E−1 17 5.78E−1 17 4.59E−2 17
#7 1.00E−5 18 6.24E−1 16 6.24E−1 16 4.80E−2 16
#8 1.03E−3 6 7.29E−1 6 7.29E−1 6 9.91E−2 5
#9 2.80E−4 11 7.16E−1 8 7.16E−1 8 7.78E−2 8 Fig. 11. Variation of the po indicator for the selected wind turbines for the site
#10 5.35E−3 2 8.01E−1 2 8.02E−1 2 1.51E−1 2
of Essaouira (see Tables 17, 18 and 19).
#11 5.00E−5 15 6.41E−1 15 6.41E−1 15 4.93E−2 15
#12 1.50E−4 13 7.11E−1 10 7.11E−1 10 6.45E−2 13
#13 3.50E−4 9 6.46E−1 14 6.46E−1 14 6.66E−2 12
#14 4.00E−4 8 6.91E−1 12 6.91E−1 12 6.89E−2 11
#15 2.91E−3 4 7.69E−1 4 7.70E−1 4 1.11E−1 4
#16 1.39E−3 5 7.59E−1 5 7.59E−1 5 9.23E−2 7
#17 4.80E−3 3 7.79E−1 3 7.80E−1 3 1.19E−1 3
#18 8.78E−3 1 8.18E−1 1 8.20E−1 1 1.52E−1 1

Fig. 12. Variation of the CF indicator for the selected wind turbines for the site
of Essaouira (see Tables 17, 18 and 19).

Fig. 9. Variation of the pfc indicator for the selected wind turbines for the site of
Essaouira (see Tables 17, 18 and 19).

Fig. 13. Wire-frame graph of ppc of wt#18 as a function of c and k measured at

given anemometer height.
Fig. 10. Variation of the ppc indicator for the selected wind turbines for the site
of Essaouira (see Tables 17, 18 and 19). estimator. The importance sampling [14] is the most efficient variance
reduction technique that favors the occurrences of negative values of
function for a representative sample (often large) from the population. the performance function. The principle of this technique consists on
These methods are the most calculation time consuming, but they are finding the design point and the use of a biased probability distribution
often used as a reference or in cases when other reliability methods fail function h (u ) i.e. importance function to bias the realization of random
to achieve a solution. The probability of failure in this case can be ex- variables around the design point. The estimator of the probability
pressed using an indicator function γf that takes 1 if G (X ) ⩽ 0 and 0 function is then expressed as follows:
otherwise: N
1 Φ(U (r ) )
N pf ≈ ∼
pf = ∑ γf (U (r ) )
1 N h (U (r ) ) (15)
pf ≈ ∼
pf = ∑ γ f(r ) r=1
N r=1 (14)

r is the rank of sampling. The convergence and accuracy of this 3.1.3. Global sensitivity analysis
method depends on the number of simulations N and the variance of the The global sensitivity analysis (GSA) is another but a

506
N. Aghbalou et al.
Fig. 14. Wire-frame graph of pfc of wt#18 as a function of c and k measured at
given anemometer height.
Fig. 15. Wire-frame graph of po of wt#18 as a function of c and k measured at
given anemometer height.
Fig. 16. Effect of α on the probability of operating in a site with k = 2.5 and
c = 6 measured 10 m AGL.
complementary study allowed by the probabilistic approach. It helps to
study the effect of the variability of stochastic variables X to the
variability of the performance function response G (X ) , by means of
indices, and hence helps on the ranking of the most influential para-
meters and decision to be made about the change in the parameters of
the model to achieve a desired reliability level. When the variables Xi
are independent, a variance based importance measures method or
Monte-Carlo based Sobol’s indices has been developed in [54]. This
method is the mostly used in the GSA as it is global, robust and doesn’t
507
Electrical Power and Energy Systems 103 (2018) 497–510
Fig. 17. First and total-order Sobol’s indices for the probability of full charge
operating based on Eq. (21).
Fig. 18. First and total-order Sobol’s indices for the probability of operating
based on Eq. (20).
require any hypothesis about the performance function [55]. To mea-
sure the individual parameter (e.g parameter x i ) contributions to the
variance of the performance function, the first-order index is expressed
as follows:
Var [E (G (X )/ x i ]
Si =
Var (G (X )) (16)
The total-order indices measure the contributions of the individual and
interactive parameters to the variance of the performance function.
n n
STi = Si + ∑ Sij + ∑ Sijk + ..
j #i j#i,k#i,j<k (17)
where Sij is the second order index measuring the sensitivity of the
model to the interaction between x i and x j , and so on [56].
3.2. Problem formulation: long term performance
As mentioned in Section 2, the produced power of wind turbine
mainly depends on the characteristics of the machine itself. Those
parameters are established by the manufacturer in some specific con-
ditions according to the procedure in [8] for external meteorological
mast or in [57] for wind turbines based on nacelle anemometry. The
other influencing factors are climatic conditions as topography and
atmospheric stability which influence the air density and the wind
shear profile. Assuming that the two operating modes of the wind
turbine (see Section 2.2) are independent, the probability of operating
po (i.e. the probability of producing electrical power) can be defined as:
po = 1−(1−pfc )(1−ppc ) (18)
where pfc the probability of operating at full charge and ppc the prob-
ability of operating at partial charge.
In structural engineering, a structure is safe in accordance with a
N. Aghbalou et al.
failure criterion: if the applied stress remains below the strength of the
component. The performance function, at any time, can be defined as
the difference between the strength of the component and the applied
stress. In our study, the produced power Pelec (.) of the WTG in Eq. (2)
can be placed in the strain side and the critical power Pc (.) in the
strength side. The performance function may be expressed as:
G (Xturbine , Xwind ) = Pelec (Xturbine , Xwind )−Pc (.) (19)
where Xturbine = (H , Vc , Vr , Vf , Pr ) is vector of random variables related
to the WTG specification and Xwind = (Vm, c, k , α, z 0) is vector of
random variables related to the wind climate.
Then, the following probabilities can be defined as follows:
pfc = 1−Pr(G1 (Xturbine , Xwind ) < 0) (20)
and
po = 1−Pr(G2 (Xturbine , Xwind ) < 0) (21)
where G1 (Xturbine , Xwind ) is the performance function as the produced
power Pelec (.) of the WTG less than or equals to zero Pc = 0 . And
G2 (Xturbine , Xwind ) is the performance function such that when the pro-
duced power Pelec (.) is less than the rated power Pc = Pr . The ppc can be
deduced from the Eqs. (18), (20) and (21).
po −pfc
ppc =
1−pfc (22)
The vector of input variables X is normally distributed around their
average value with a coefficient of variance equals to 2%. Except the
height at which we attempts to install the wind turbine and the rated
power of the wind turbine is normally distributed around the mean
value with a very small coefficient of variance equals to 10−7% and
10−4% respectively. The probabilities are performed through FORM
method. Wind speed data for this study was collected in two sites in
Morocco. Table 6 shows the information of the collected data. The data
recorded every 1 h from January (1st, 2015 to February 29th, 2016.
Table 5 lists the power curve parameters of 18 wind turbines selected
for this study.
4. Results and discussion
4.1. The comparison between the Weibull parameters estimators
The wind speed measurements at two studied sites have been fit to
the Weibull probability density function (part 1 of Fig. 4). The statis-
tical parameters c, k and mean wind speed Vm have been estimated
using selected methods LSE , MLE , MaxEnt1, MaxEnt 2, CS and PSO are
given in Tables 7 and 8. Based on the Root Mean Square Error (RMSE)
goodness-of-fit, the PSO algorithm performs slightly better than the CS
method, but well better than other estimators in the site of Dakhla. The
same result is obtained in Essaouira site. It is well to mention that the
MaxEnt1 shows a good performance compared to MaxEnt2, LSE and
MLE. Hence, the PSO algorithm is selected to be used to estimate
Weibull parameters for extrapolated wind speed in the next.
4.2. Wind power assessment at the mast hub height
In this subsection, the paper investigates the variation of CF and the
proposed probabilities, by assessing the effectiveness of hypothetical
wind turbine Enercon E-33 (H = 50 m, Vc = 3 m/s, Vr = 12.5 m/s,
Vf = 25 m/s and Pr = 330 m/s) at the height of the measured wind
speed H = 50 m in the studied sites, considering the different estimators
and models as described in Fig. 4 part 2. Tables 9 and 10 present a
summary of the calculated probabilities and capacity factors for the
hypothetical wind turbine. Results reveal that the probability of oper-
ating po is mostly influenced by the probability of operating at the
partial charge ppc . These probabilities are approximately the same for
508
Electrical Power and Energy Systems 103 (2018) 497–510
the different models ( g1 (v ), g2 (v ), g3 (v ) and g4 (v ) ). However, these
change with the methods used for statistical parameter estimation of
Weibull probability density function. For example, in the site of Dakhla
ppc /CS for models g1 (v ), g2 (v ), g3 (v ) and g4 (v ) is about 9.75E−1. In the
site of Essaouira po /PSO is 8.51E−1 for g1 (v ), g2 (v ), g3 (v ) . However, the
capacity factor values vary significantly with models. For example CF/
PSO varies from 2.24E−1 to 4.02E−1 in the site of Dakhla. The CF/
PSO varies from 1.51E−1 to 3.02E−1 in the site of Essaouira. More-
over, based on the values of standard deviation σ , it is observed that po
is more sensitive to the Weibull parameter variation as values of po are
more dispersed around the average values comparing to the CF values,
except for model g3 (v );CF seems to be more sensitive values compared
to po . Table 11 presents a descriptive statistics; mean and standard
deviation of each indicators po and CF when changing the models
( g1 (v ), g2 (v ), g3 (v ), g4 (v ) and g5 (v ) ) for the same estimator
(LSE , MLE …), and Table 12 when changing the estimator of the Wei-
bull wind speed distribution parameter’s. Based on the values of stan-
dard deviation σ , the values of CF are more dispersed around the
average values comparing to the po values. Then, the probability of
operating po is less sensitive to the variation of the model used to ap-
proximate the power curve comparing to CF.
4.3. Site- turbine matching
The estimated Weibull parameters, presented in Table 13, are used
to calculate the probabilities and the capacity factors for each wind
turbine in Table 5 and for each model in Table 1 as described in Fig. 4
part 3. The surface roughness length is assumed equal to z 0 = 0.03 m
based on the expert knowledge of the two sites. Then, the wind shear is
calculated based on Eq. (6) and the wind speed is extrapolated to the
hub height of each wind turbine using Eq. (5). In Tables 14, 15 and 16
wind turbines are ranked according to the different probabilities and
capacity factors for the site of Dakhla. The variation of the calculated
indicators pfc , ppc , po and CF for this site are graphically depicted in
Figs. 5, 6, 7 and 8. According to the CF indicator, wt#18, wt#10, wt#17,
wt#15 and wt#16 are the 1st, 2nd, 3rd, 4th and 5th match respectively
for this site based on models g2 (v ), g3 (v ), g4 (v ) and g5 (v ) . It is worth-
while to note that wt#18 has relatively the lowest Vf = 20 m/s. How-
ever, based on model g1 (v ) , the 1st, 2nd, 3rd, 4th and 5th match are
wt#10, wt#17, wt#1, wt#16 and wt#13 respectively; these wind tur-
bines have the same Vf = 25 m/s. wt#10 has the lowest Vr = 12.5 m/s.
According to the po indicator, wt#13, wt#14 and wt#1 are the best, 2nd
and 3rd match for this site based respectively on models g1 (v ), g2 (v ) and
g3 (v ) . However, wt#13, wt#1 and wt#14 are the 1st, 2nd and 3rd match
based on model g4 (v ) . These wind turbines have the same Vf = 25 m/s ,
but wt#14 have the highest Vc = 3.5 m/s and wt#1 have the lowest
Vr = 14 m/s . The wt#16 and wt#10 are respectively the 4th and 5th for
models g2 (v ), g3 (v ) and g4 (v ) . Based on model g5 (v ) , the 1st, 2nd, 3rd,
4th and 5th match are wt#16, wt#10, wt#12, wt#8 and wt#15 re-
spectively. These wind turbines have the same Vf = 25 m/s, wt#10
have the lowest Vr = 12.5 m/s. In Tables 17, 18 and 19 wind turbines
are ranked according to the different probabilities and capacity factor
for the site of Essaouira. The variation of the calculated indicators pfc ,
ppc , po and CF for this site are graphically depicted in Figs. 9, 10, 11 and
12. For this site, wt#17 is the best according to po , the 2nd, 3rd, 4th and
5th match are respectively wt#18, wt#12, wt#13 and wt#1 based on
models g1 (v ), g3 (v ) and g4 (v ) . wt#18, wt#12, wt#13 and wt#1 are the
2nd, 3rd, 4th and 5th match are respectively based on g2 (v ) . wt#17
have the highest Vr = 14 m/s and Vf = 25 m/s comparing to wt#18.
wt#17 and wt#1 have the same characteristics (Vc , Vr and Vf ) but pof is
highest for wt#17, thus wt#17 has the highest hub height. However,
wt#18, wt#10, wt#17, wt#15 and wt#1 are the best, 2nd, 3rd, 4th and
5th match respectively according to CF based on models g2 (v ), g3 (v ) and
g4 (v ) . Based on model g5 (v ) , CF and po indicator get different ranking
for these studied sites. Based on Tables 14, 15, 16, 17, 18 and 19, some
remarks are drawn:
N. Aghbalou et al.
Ranking according to CF is almost similar to pfc . In fact, these two
indicators are engaged to assess the rate of operating at the rated
power.
pfc is often smaller than the ppc , thereby it influences slightly the po .
It is important to take this information into consideration when de-
veloping a wind farm mostly when optimizing the capture of energy by
means of MPPT devices needed.
According to po and ppc , for sites with small shape parameter, the
highest wind turbine is not desired (case of Dakhla site).
4.4. Effect of the wind shear exponent
As mentioned above, the wind shear profile varies from site to site,
depends on the surface ground roughness. It fluctuates with the atmo-
spheric stability and by seasonal and diurnal changes. This subsection
studies the effect of wind shear po as described in Fig. 4 part 4. Fig. 16
illustrates the variation of the po of different wind turbine in Table 5 for
the same wind characteristics (k = 2.5 and c = 6 measured at 10 m
AGL). It is observed that for each wind turbine the effect of α is dif-
ferent. Nevertheless, all wind turbines the po increases simultaneously
with the increase of α . Furthermore, it can be noticed that the effect is
important for turbines with higher cut-in speed. For example, wt#6
with the higher cut-in speed in the selected wind turbines Vc = 5 m/s
shows a great sensitivity to the variation of α , however for wind tur-
bines wt#1, wt#10 wt#13, wt#17 and wt#18 with lower cut-in speed
Vc = 3 m/s the effect is less significant. The result reveals that for wind
turbines with the same cut-in speed and tower heights; wt#2 with wt#3
and wt#4 or wt#8 with wt#9 or wt#17 with wt#18; for all wind shear
component α scenarios the probability of operating po is the same.
Another interesting remark, is for wind turbine with the same rated
speed Vr = 16 m/s and cut-out speed Vf = 25 m/s, for all wind shear
component α scenarios, the effectiveness of the wind turbine doesn’t
increase with the tower height h and the rated power Pr .
4.5. Effect of Weibull parameters
Figs. 13–15 show the ware-frame graphs of the proposed prob-
abilities ppc , pfc and po for hypothetical wind turbine (e.g. wt#18) for a
range of Weibull shape parameter from 1.5 to 10 and scale parameter
from 0 to 25 m/s (see Fig. 4 part 5). The power curve of this wind
turbine is supposed modeled by the quadratic model g2 (v ) . Based on
these curves, it is observed that the probability of operating at the full
charge pfc increases with scale parameters and starts reaching the
maximum values for shape parameter between 2 and 3 and scale
parameter between 9 m/s and 10 m/s. This means that the probability
of full charge operating pfc is mostly influenced by the mean wind speed
as this increases with scale parameter [3]. The probability of the partial
charge ppc operating is relatively less influenced by the scale parameter
than the shape parameter. This parameter is location specific and
doesn’t vary with height as the shape parameter. Then, the probability
of the partial charge operating informs better on the profitability of the
selected wind turbine for the lowest wind speeds. It is worth men-
tioning that for sites with scale parameter upper than 1.5, the prob-
ability of the partial charge operating ppc of this wind turbine reaches
the maximum for shape parameter between 7.5 m/s 8.5 m/s and then
decreases.
4.6. Global sensitivity analysis
Nine input uncertain parameters are used for the sensitivity analysis
of the studied performance functions Eqs. 20 and 21 as described in
Fig. 4 part 6. The first-and total-order sensitivity indices of these
parameters are shown in Figs. 17 and 18 for wind speed at h = 100 m
and wt#18 in the site of Dakhla. The sensitivity analysis reveals an
obvious significance of the wind speed parameter as its total sensitive
index exceeds 99% for the both performance functions Eqs. 20 and 21.
509
Electrical Power and Energy Systems 103 (2018) 497–510
Additionally, it can be seen from these figures that the rated wind speed
is also a sensitive parameter compared to other parameters. It is worth
to note that the same result was observed when analyzing these per-
formances for other wind turbines with the corresponding wind speed
distribution.
5. Conclusion
The wind turbine-site matching in wind power engineering is based
on some deterministic indicators as for example the capacity factor. The
objective is often ranking wind turbines according to their effectiveness
considering their power curve performance and probability distribution
of measured or extrapolated wind speed in the selected site. All in-
volved parameters are assumed deterministic, methods and models are
selected based on recommendation from literature and expert judg-
ment. However, the random behavior of wind speed climate and the
uncertainties of Wind Turbine Generator characteristics are important
to take into account in models used to evaluate the WTG performance.
The impact of the uncertainty on a single wind turbine performance can
have a significant consequence in terms of annual energy production
and then the profitability of a wind power plant project.
After having reviewed the wind power modeling process, a for-
mulation of the proposed probabilistic reliability performance based
method has been introduced and applicability has been investigated.
The uncertainty can arise from wind speed modeling as highlighted by
testing six statistical methods recommended in literature. The un-
certainty can arise also from the power curve modeling as highlighted
by testing five commonly used power curve approximation models. The
conclusions are summarized as follows:
1. The selected methods for Weibull parameters estimation are com-
pared in terms of RMSE to estimate the distribution function of wind
speed at the measured height in the studied sites. The results re-
vealed that each method returns a different value and the PSO al-
gorithm achieves the best result.
2. The variability of the proposed performance indicators and the
conventional indicator CF to the variation of Weibull parameters
and to the different models for power curve modeling have been
identified by assessing the effectiveness of a wind turbine selected
with same tower hub height as the height of the measured wind
speed. Results revealed that the introduced method is less sensitive
to the models used to approximate the wind power curves compared
to the conventional method.
3. A turbine-site ranking based on this method and the conventional CF
indicator has been achieved using wind turbines with different
tower height and characteristics. Results revealed that the method is
less sensitive to the variation of the models used in the power curve
approximation and leads to the same ranking. However, the con-
ventional CF gets different rankings which may lead to problems of
over-sizing or under-sizing.
4. The effect of the Weibull parameters has been studied for a hy-
pothetical wind turbine to assess the variation of the proposed
probabilities with respect to these parameters at the anemometer
height. Result shows that this probabilistic approach can inform
better about the profitability of a selected wind turbine for each
operating mode at any selected site.
5. The effect of the wind shear variation on the turbine-site ranking
problem has been investigated. Results show that this effect varies
with respect to the wind turbine characteristics itself.
6. A sensitivity analysis has highlighted that the wind speed and the
rated wind speed are the most significant parameters with regards to
the variability of the proposed performance functions responses.
The modeling of the stochastic variables was beyond the scope of
this study. It might be conducted based on a fine database; regional or
international measured performance of the wind turbine or the wind
N. Aghbalou et al.
shear, since the objective of this study was to propose and investigate a
probabilistic framework to study the problem turbine-site appraising in
an uncertain framework.
References
[1] Albadi MH, El-Saadany EF. New method for estimating CF of pitch-regulated wind
turbines. Electric Power Syst Res 2010;80(9):1182–8.
[2] Lee J, Kim DR, Lee KS. Optimum hub height of a wind turbine for maximizing
annual net profit. Energy Convers Manage 2015;100:90–6.
[3] Albadi MH, El-Saadany EF. Optimum turbine-site matching. Energy
2010;35(9):3593–602.
[4] Dong Yao, et al. Intelligent optimized wind resource assessment and wind turbines
selection in Huitengxile of Inner Mongolia, China. Appl energy 2013;109:239–53.
[5] Chang TP, et al. Site matching study of pitch-controlled wind turbine generator.
Energy Convers Manage 2014;86:664–9.
[6] Sainz E, Llombart A, Guerrero JJ. Robust filtering for the characterization of wind
turbines: Improving its operation and maintenance. Energy Convers Manage
2009;50(9):2136–47.
[7] Kusiak A, Zheng H, Song Z. Models for monitoring wind farm power. Renew Energy
2009;34(3):583–90.
[8] International Electrotechnical Commission et al. Part 12–1: power performance
measurements of electricity producing wind turbines; IEC TC/SC 88. Tech. rep. IEC
61400-12-1; 2005.
[9] Jung S, Arda VO, Kwon SD. Wind energy potential assessment considering the
uncertainties due to limited data. In: Applied Energy. Special Issue on Advances in
sustainable biofuel production and use – XIX International Symposium on Alcohol
Fuels – ISAF 102 (Feb. 2013). p. 1492–503.
[10] Melchers RE. Structural reliability: analysis and prediction. Horwood; 1987.
[11] Charki A, et al. Reliability of a hydrostatic bearing. J Tribol 2014;136(1).
[12] Charki A, et al. Lifetime assessment of a photovoltaic system using stochastic Petri
nets. Int J Model Simul 2017;37(3):149–55.
[13] Titikpina F, et al. Uncertainty assessment in building energy performance with a
simplified model. Int J Metrol Quality Eng 2015;6(3).
[14] Stewart M, Melchers RE. Probabilistic risk assessment of engineering systems.
Springer; 1997.
[15] Ditlevsen O, Madsen HO. Structural reliability methods vol. 178. New York: Wiley;
1996.
[16] Wang Jianzhou, Jianming Hu, Ma Kailiang. Wind speed probability distribution
estimation and wind energy assessment. Renew Sustain Energy Rev
2016;60:881–99.
[17] Ouarda TBMJ, et al. Energy Convers Manage 2015;93:414–34.
[18] Ouarda TBM J, Charron C, Chebana F. Review of criteria for the selection of
probability distributions for wind speed data and introduction of the moment and L-
moment ratio diagram methods, with a case study. Energy Convers Manage
2016;124:247–65.
[19] Carrillo C, et al. Review of power curve modelling for wind turbines. Renew Sustain
Energy Rev 2013;21:572–81.
[20] Seguro JV, Lambert TW. Modern estimation of the parameters of the Weibull wind
speed distribution for wind energy analysis. J Wind Eng Ind Aerodyn
2000;85(1):75–84.
[21] Wais P. Two and three-parameterWeibull distribution in available wind power
analysis. Renew Energy 2017;103:15–29. ISSN: 0960-1481.
[22] Albadi MH, El-Saadany EF, Albadi HA. Wind to power a new city in Oman. In:
Energy. 11th conference on process integration, modelling and optimisation for
energy saving and pollution reduction 34 (10); 2009. p. 1579–86.
[23] Azad AK, Rasul MG, Yusaf T. Statistical diagnosis of the best Weibull methods
forwind power assessment for agricultural applications. Energies
2014;7(5):3056–85.
[24] Liu FJ, et al. Wind characterization analysis incorporating genetic algorithm: a case
study in Taiwan Strait. Energy 2011;36(5):2611–9.
[25] Justus CG, Mikhail A. Height variation of wind speed and wind distributions sta-
tistics. Geophys Res Lett 1976;3(5):261–4.
[26] Zhang H, Yu YJ, Liu ZY. Study on the maximum entropy principle applied to the
annual wind speed probability distribution: a case study for observations of inter-
tidal zone anemometer towers of Rudong in East China Sea. Appl Energy
2014;114:931–8.
[27] Zellner A, Highfield RA. Calculation of maximum entropy distributions and
510
Electrical Power and Energy Systems 103 (2018) 497–510
approximation of marginalposterior distributions. J Econom 1988;37(2):195–209.
[28] Shamilov A, Kantar YM, Usta I. Use of MinMaxEnt distributions defined on basis of
MaxEnt method in wind power study. Energy Convers Manage 2008;49(4):660–77.
[29] Yang XS, Deb S. Cuckoo Search via Lévy flights. In: 2009 world congress on nature
biologically inspired computing (NaBIC). Dec. 2009. p. 210–4.
[30] Kennedy J. Particle swarm optimization. Encyclopedia of machine learning.
Springer; 2011. p. 760–6.
[31] Lydia M, et al. A comprehensive review on wind turbine power curve modeling
techniques. Renew Sustain Energy Rev 2014;30:452–60.
[32] Sohoni V, Gupta SC, Nema RK. A critical review on wind turbine power curve
modelling techniques and their applications in wind based energy systems. J Energy
2016;2016. e8519785.
[33] Gill S, Stephen B, Galloway S. Wind turbine condition assessment through power
curve copula modeling. IEEE Trans Sustain Energy 2012;3(1):94–101.
[34] Thapar V, Agnihotri G, Sethi VK. Critical analysis of methods for mathematical
modelling of wind turbines. Renew Energy 2011;36(11):3166–77.
[35] Yang HX, Lu L, Burnett J. Weather data and probability analysis of hybrid photo-
voltaicwind power generation systems in Hong Kong. Renew Energy
2003;28(11):1813–24.
[36] Hu SY, Cheng JH. Performance evaluation of pairing between sites and wind tur-
bines. Renew Energy 2007;32(11):1934–47.
[37] Akorede MF, et al. Appraising the viability of wind energy conversion system in the
Peninsular Malaysia. Energy Convers Manage 2013;76:801–10.
[38] Yeh TH, Wang L. A study on generator capacity for wind turbines under various
tower heights and rated wind speeds using Weibull distribution. IEEE Trans Energy
Convers 2008;23(2):592–602.
[39] Huang SJ, Wan HH. Determination of suitability betweenwind turbine generators
and sites including power density and capacity factor considerations. IEEE Trans
Sustain Energy 2012;3(3):390–7.
[40] Jangamshetti SH, Rau VG. Normalized power curves as a tool for identification of
optimum wind turbine generator parameters. IEEE Trans Energy Convers
2001;16(3):283–8.
[41] EL-Shimy M. Optimal site matching of wind turbine generator: case study of the
Gulf of Suez region in Egypt. Renew Energy 2010;35(8):1870–8.
[42] AbulWafa Ahmed R. Matching wind turbine generators with wind regime in Egypt.
Electric Power Syst Res 2011;81(4):894–8.
[43] Tong W. Wind power generation and wind turbine design. Google-Books-ID:
wU9bgvrl4rQC. WIT Press; 2010. p. 769.
[44] Toft HS, et al. Uncertainty in wind climate parameters and their influence on wind
turbine fatigue loads. Renew Energy 2016;90:352–61.
[45] Schallenberg-Rodriguez J. A methodological review to estimate techno-economical
wind energy production. Renew Sustain Energy Rev 2013;21:272–87.
[46] Aghbalou N, et al. Horizontal extrapolation of wind speed distribution using neural
network for wind resource assessment. Int J Sci Res 2017;6.
[47] Aghbalou N, et al. A Bayesian regularized artificial neural network for up-scaling
wind speed profile. Int J Curr Eng Technol 2017;5.
[48] Spera DA, Richards TR. Modified power law equations for vertical wind profiles.
Conference and workshop on wind energy characteristics and wind energy siting:
Portland, OR, USA; June 1979.
[49] Verma AK, Ajit S, Karanki DR. Reliability and safety engineering. Google-Books-ID:
OgWhCgAAQBAJ. Springer; 2015. p. 583. ISBN: 978-1-4471-6269-8.
[50] Troels Friis P et al. Wind turbine power performance verification in complex terrain
and wind farms. Report 87-550-3028-9; 2002.
[51] Pelletier F, Masson C, Tahan A. Wind turbine power curve modelling using artificial
neural network. Renew Energy 2016;89:207–14.
[52] Masters GM. Renewable and efficient electric power systems. John Wiley & Sons;
2013.
[53] Nelson V. Wind energy: renewable energy and the environment. CRC Press; 2013.
[54] Homma T, Saltelli A. Importance measures in global sensitivity analysis of non-
linear models. Reliab Eng Syst Safety 1996;52(1):1–17. ISSN: 0951-8320.
[55] Iooss B, Lemâtre P. A review on global sensitivity analysis methods. In: Uncertainty
management in simulation-optimization of complex systems. In: Dellino Gabriella,
Meloni Carlo, editors. Operations research/computer science interfaces series 59.
US: Springer; 2015. p. 101–22.
[56] Sobol IM. Sensitivity estimates for nonlinear mathematical models. Math Modell
Comput Exp 1993;1:40714.
[57] International Electrotechnical Commission et al. Power performance of electricity
producing wind turbines based on nacelle anemometry. Tech. rep. IEC 61400-12-2
CD Part 12-2; 2008.