Renewable and Sustainable Energy Reviews 58 (2016) 1048–1059

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews

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

Wake effect modeling: A review of wind farm layout optimization

using Jensen's model
Rabia Shakoor a, Mohammad Yusri Hassan a,n, Abdur Raheem a,b, Yuan-Kang Wu c
a
Centre of Electrical Energy Systems (CEES), Faculty of Electrical Engineering, Universiti Teknologi Malaysia (UTM), 81310 Johor Bahru, Johor, Malaysia
b
University College of Engineering & Technology (UCE&T), The Islamia University of Bahawalpur (IUB), 63100 Bahawalpur, Pakistan
c
Department of Electrical Engineering, National Chung-Cheng University, 62102 Chiayi, Taiwan

art ic l e i nf o a b s t r a c t

Article history: The layout scheme of wind farms is a challenging job having many design objectives and constraints due
Received 19 June 2014 to the multiple wake phenomenon. Wake effect calculation is one of the significant problems in the wind
Received in revised form farms and needs to be modeled to decrease the power loss due to near and far wake effect. This paper
5 December 2015
reviews the wake models in general and far wake models in particular. The comparison of different far
Accepted 24 December 2015
wake models shows that the Jensen's far wake model is a good choice to solve the wind farm layout
problem due to its simplicity and relatively high degree of accuracy. This research also focuses on the
Keywords: studies carried out on wind farm layout optimization problem and the current state of the art of fitness
Wind farm functions used for the optimization of wind farms using Jensen's wake model. It is found that there is a
Wake effect
need of more optimization techniques to be applied to solve the layout problem. In addition, future
Jensen's wake model
advancements have been identified for better positioning of wind turbines in larger wind farms.
Layout optimization
& 2016 Elsevier Ltd. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1048
2. Wake modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1050
2.1. Kinematic model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1050
2.1.1. Jensen's wake model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1050
2.1.2. Larsen's model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1052
2.1.3. Frandsen's model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1053
2.2. Field models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1053
2.2.1. Eddy viscosity/2D field models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1053
2.2.2. Three dimensional field models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1053
3. Comparative study of different wake models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1053
4. Wind farm layout optimization using Jensen's wake model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1054
5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1057
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1058

1. Introduction biogas, solar etc. [1]. One of the prolific sustainable and renewable
source of energy is the wind energy [2,3]. Wind energy installation
Depletion of the fossil fuels has led to acute scarcity of energy has experienced a tremendous increase in the past decade. At the
production from the conventional source, promoting an upsurge in same time, related research activities have flourished [4]. The
utilization of the non-conventional energy resources like wind, Global Wind Energy Council 2014 Report [5] stated that wind
energy has become the most rapid rising source of energy in the
world, having a steep increase in development from 2009 to date,
n
Corresponding author. Tel.: þ 60 75557001; fax: þ 60 75557005. as shown in Fig. 1. The global installed wind capacity from 1997 to
E-mail address: [email protected] (M.Y. Hassan). 2014. In 2004 the total worldwide wind capacity was 14,781 MW

http://dx.doi.org/10.1016/j.rser.2015.12.229
1364-0321/& 2016 Elsevier Ltd. All rights reserved.
 R. Shakoor et al. / Renewable and Sustainable Energy Reviews 58 (2016) 1048–1059 1049

Fig. 1. Global annual installed capacity from 1996 to 2014 [5].

but in 2014 the capacity became 51,477 MW. Due to rapid devel- turbines are exposed to low wind speeds and high turbulence
opment of wind turbine technology and increasing size of wind intensities inside a wake region. Low wind speeds mean low
farm; starting from 4 GW in the current construction to 40 GW by energy production in wind farms, and high turbulence intensities
2020, and 150 GW by 2030, this means many large offshore wind lead to a high level of velocity fluctuation. Thus, it is necessary for
farms will be built. Wind power plays a significant part in the wind farm developers to fully understand that altering the wind
power production in developing countries as well as in developed farm layout affects not only wake losses but also the power pro-
countries [6–13]. As a result of this cumulative usage of wind duction of wind farm. Therefore, it is very important to optimize
resource, large number of wind turbines are installed in cluster the wind turbine positions in the wind farm to decrease the
form called ‘Wind farm’. The main task of a wind farm is to get as ambiguity in output power [29]. The optimization done by chan-
much energy as possible from minimal number of wind turbines
ging the position of one turbine or the other known as Wind Farm
and with a minimal space between the turbines.
Layout Optimization (WFLO). This optimization problem consists
The conversion of wind energy into useful energy involves two
of the design of wind turbine layout, which is subjected to various
processes: firstly is extracting the kinetic energy from wind and
financial and engineering objectives and constraints [30,31]. Lay-
conversion into mechanical energy at the rotor axis, and secondly
out optimization problem can be solved by using available com-
is the conversion into useful energy [14]. In the primary process,
wind turbines extract the energy from wind, and reduce the wind mercial software packages. These tools require the input from the
speed behind the rotor and swirl the air flow, which is known as user to get the improved layout projects than only produced by
wake effect of wind turbine. Thus, the downstream wind turbines the software without the human intervention [32].
receive a modified wind inflow both in terms of mean velocity and The wind farm layout design is categorised as an NP-hard
turbulence, producing less energy. The wind farm wake effect is an optimization problem containing large number of constraints. Due
important issue during the whole life span of a wind farm. The to excessive computation time, the exact algorithms would flop.
more advance wind turbines are becoming larger. Therefore, it is Because of the intricacy of layout optimization of wind farm, rig-
very important to understand the aerodynamic nature and the orous optimization approaches such as branch-and-bound,
properties of wake effects in order to calculate the optimal wind dynamic programming, backtracking and linear programming,
farm structure with the new models of wind turbines for provid- etc., can be utilized to some extent [33]. Therefore, there is a need
ing the maximal energy yield. The correct study of the wake effect to use the most commonly used algorithm of meta heuristic
can ensure the capability to control and adjusting the shadowing optimization except in situations of small wind farms [34]. There
of inner wind turbines to decrease the wake loss and to increase are a number of optimization techniques which have been suc-
the energy yield. cessfully used in wind farm layout problem, such as Genetic
In order to reduce this negative effect on the wind farm, wake Algorithms [35,36], Simulated Annealing [37,38], Differential
deficits need to be modeled to find an optimal situation. There has Evolution (DE) [39], Simulated Evolution (SimE) [40,41], Ant Col-
been a number of models developed aiming for a better under- ony Optimization (ACO) [42], Particle Swarm Optimization (PSO)
standing on wake dynamics. These models can be divided into two [43], Stochastic Evolution [44,45], Definite Point Selection (DPS),
main categories, namely analytical/empirical/explicit wake models
bionic optimization, Gradient based optimization, numerical
and computational/implicit wake models. An analytical wake
added simulation and monte carlo optimization technique. An
model characterizes the velocity in a wake through a set of ana-
optimization task often begins with setting a model of a set of
lytical expressions. These are based on the conservation of mass
mathematical equations called objective functions and constraints
and empirical relations of wake decay, which are mainly used for
that describe the problem at hand. Ideally, this model should
micro siting and wind farm output predictions. These models
accurately reflect the physics of the problem and also be easily
attempt to characterize the energy content in the flow field and
ignore the details of the exact nature of the flow field [15–22]. solvable.
Meanwhile, in computational wake models, fluid flow equations, This review highlights the status of wake effect in the wind
whether simplified or not, must be solved to obtain the wake farm and discusses its significance on the energy yield. In wind
velocity field [23–28]. farm layout optimization problem, the far wake effect is more
Wind energy assessment is a game of inches, where percentage important than near wake effect. Primarily, the present research
points mean differences in tens of millions of dollars. Modern describes and compares some far wake models from the literature.
wind farms consist of tens of wind turbines arranged on the sites The next section presents research on wind farm layout optimi-
with the purpose of maximum utilization of wind energy. How- zation using Jensen's wake effect model. Finally, the concluding
ever, the aerodynamic behavior of wind turbines will generate remarks include an analysis of advance research prospects and
large scale wakes in the downwind field, so the downwind defies.
1050 R. Shakoor et al. / Renewable and Sustainable Energy Reviews 58 (2016) 1048–1059

2. Wake modeling

Wind turbine wakes are classified into two type according to

power losses and loading, which are near wake and far wake. The
near wake is the region from the turbine to almost two to three
rotor diameters downstream, where the turbine geometry directly
disturbs the wind [46]. Fig. 2 shows the velocity profile behind the
wind turbine. The far wake is the region past the near wake, where
the emphasis is on the impact of wind turbines in large farm
conditions [47]. The shear layer is a velocity difference between
the air outside and inside of the wake, which expand towards
downstream. Turbulence is the more dominating factor in the far
wake region than in near wake region. In the far wake, two main Fig. 2. The velocity profile behind wind turbine [47].
mechanisms defining flow conditions are convection and turbu-
lent diffusion. A parabolic approximation is applicable to manip-
ulate this region in many conditions. It is expected that sufficiently wake expands linearly downstream. If the near field behind a wind
far downstream, the damaging effects of momentum deficit and turbine is neglected, the resulting wake behind the wind generator
increased level of turbulence will disappear because of turbulent can be treated as a turbulent wake. This model is based on the
diffusion of the wake. Although it is difficult to separate both wake assumption that the wake is a turbulent and the contribution of tip
effects since near wake effect becomes the initial conditions of far vortices is neglected. Thus, this means that this wake model is
wake effect. In wind turbine optimization problem, the far wake strictly applicable only in the far wake region. Vanluvanee [50],
becomes more important than near wake. The simplest method is based on the findings from his study, recommends the N.O. Jen-
to solve the problem analytically by exploiting the self-similar sen's model to be used for the energy predictions in offshore wind
nature of the far wake to get formulation for the turbulence farms, as it gives a good trade-off between prediction errors.
intensity and the velocity deficit.
Far wake models are basically divided into two groups namely 2.1.1.1. Jensen single wake model. The Jensen's wake model is
Kinematic models [23], Field and Wake edded turbulence model derived by conserving momentum across a control volume in the
models [26,57,58], as shown in Fig. 3. Kinematic wake models are wind turbine's wake. The law of conservation of momentum
based on self-similar velocity profile. The Larsen wake model [59], implies that the radius of wake (r) behind the turbine expands
Frandsen [22] and Jensen's wake model [19] belong to this class. linearly with respect to downwind distance (D) [66]. Fig. 4 shows
For the wake measurements the Field and Wake edded turbulence the wake effect behind a wind turbine facing free stream wind
models solve the Reynolds-averaged Navier–Stokes equations with velocity vo , where v is the wake wind speed just behind the wind
a turbulence model. These are further divided into Eddy Viscosity turbine, r o is the rotor radius of wind turbine, r and v1 represent
or two dimensional [26] and three dimensional field models [31]. the wake radius and wake velocity respectively at down wind
distance x.
2.1. Kinematic model Eq. (1) gives the expression for law of conservation of
momentum for wind turbines.
The first method to study wind turbine wakes was introduced
in a seminar paper by Lissaman [22], known as kinematic model. π r2o v þ π ðr2  r o 2 Þvo ¼ π r2 v1 ð1Þ
Kinematic models are based on self-similar velocity deficit profiles
obtained from comparison of academic and tentative work on
To get the relationship of downstream velocity, Eq. (1) can be
coflowing jets. Kinematic wake models employ only the momen-
solved with respect to velocity. The resulting equation gives the
tum equation to model the velocity deficit of the wake behind a
downstream velocity as a function of upstream wind velocity.
turbine. Besides that, the wake descriptions do not consider the
According to Betz theory the value of v in terms of vo will be given
initial expansion region of the wake. They also do not cover the
as in the following equation [67]:
change in turbulence intensity in the wake behind a turbine, thus
they have to be coupled with a turbulence model if the values of v ¼ ð1  2aÞvo ð2Þ
the turbulence intensity in the wakes and throughout the wind
where a is the axial flow induction coefficient or induction factor.
farm are desired or when used for load calculations [48]. The
It is assumed that the wake downstream expands linearly. Thus,
Kinematic wake models have to be combined with turbulence
the path tracing by the wind downstream follows the conical
models when used for load calculation. In most softwares, the
shape of disturbance. The radius of the cone can be estimated by
turbulence models available are Larsen, the Frandsen or DIBt
using the following equation:
model [49] and the Quarton/TNO turbulence model.
RW ¼ dð1 þ 2αxÞ=2 ð3Þ
2.1.1. Jensen's wake model
where d is wind turbine diameter and Rw is radius of
One of the oldest and most widely used wake model was
expanding wake.
developed originally by N.O. Jensen [15] (now referred to as the
The dimensionless scalar α, known as decay constant, deter-
park model), which was modified by Katic [17]. It is quite a simple
mines how quickly the wake expands with distance or describes
wake model, assuming a linearly expanding wake with a velocity
the growth of the wake width. The determination of α is sensitive
deficit that is only dependent on the distance behind the rotor.
to factors including ambient turbulence, turbine induced turbu-
Jensen treated the wake behind the wind turbine as a turbulent
lence and atmospheric stability. This can be calculated by using an
wake which ignores the contribution of vortex shedding that is
analytical expression [15], as below Eq. (4):
significant only in the near wake region. The wake model is, thus,
derived by conserving the momentum downstream of the wind 0:5
α¼
 ð4Þ
turbine. The velocity in the wake is given as a function of down- ln z
zo
stream distance from the turbine hub and it is assumed that the
 R. Shakoor et al. / Renewable and Sustainable Energy Reviews 58 (2016) 1048–1059 1051

Fig. 3. Far wake models.

Fig. 5. Multiple wake effect in wind farm.

distance, the velocity profile is called ‘hat-shaped’. It can easily be

seen from Eq. (6) that just behind the rotor s ¼ 0 the velocity in the
Fig. 4. Jensen's single wake model [51]. wake will be expressed as.
pffiffiffiffiffiffiffiffiffiffiffiffiffi
uj s ¼ 0 ¼ U 1 1  C T ð7Þ

where, z is the wind turbine's hub height and, zo represents the

surface roughness of wind farm area. The parameter zo is crucial in Hence due to its simplifications, the Jensen's model requires
the decay coefficient calculation. The decay constant has a default that the thrust coefficient of the rotor is smaller than one.
value of 0.075 in most land cases. For offshore applications, it is Although simple, this vigil shape model has been shown to accu-
recommended to use an α value of 0.04. rately calculate the coil celerity reduction in the greatly a wake
The Eq. (5) describes the wind velocity inside the single turbine inclose [53,54]. The Jensen's model is the wake model in WAsP
wake area, which is the basic approach of Jensen’s single wake (Wind Atlas Analysis and Application Program), and is also inclu-
model. However, the velocity function is not accurate in the near ded in Garrad Hassan (GH) WindFarmer software. Until the
wake regime; it can only be used in a range of far wake regions, upgrade to version 2.5, it has also been the only velocity deficit
which is approximately 3–5D on-shore or 6–8D off-shore [52]. model available in Wind PRO. This modified version allows the
pffiffiffiffiffiffiffiffiffiffiffiffiffi
r 2 Jensen's model to be able to work together with turbulence
v1 ¼ vo þ vo ð 1  C T  1Þ d ð5Þ models. Which enable this model for the predictions of near wake.
r
The Farm Layout Program (FLaP) [55] was developed by the
University of Oldenburg (UO) and is also based on the Jensen's
The velocity in the (fully developed) wake is given by Eq. (6): wake model.
" pffiffiffiffiffiffiffiffiffiffiffiffiffi#
1  1  CT
v1 ¼ vo 1  π r2 ð6Þ 2.1.1.2. Jensen's multiple wake model. The wake effect becomes
ð1 þ 2αsÞ2
more severe when the number of wind turbines in a wind farm
increasing or even in the neighboring wind farms. One target
The wake radius r and velocity v1 both depend on the relative turbine may be affected by the wakes of more than one turbine.
distance D behind the rotor, s ¼ D=2r and C T is thrust coefficient of This situation is called multiple wake effect [56]. Fig. 5 shows the
upwind turbine and it is a function of the induction factor. Because multiple wake effect of a wind farm having 7 wind turbines. The
the velocity in the wake is constant for a given downstream wind turbines T1, T2, T3 are facing free-stream velocity while the
1052 R. Shakoor et al. / Renewable and Sustainable Energy Reviews 58 (2016) 1048–1059

Fig. 6. Wind turbine wake shadow [51].

wind turbines T4, T5, T6 are operating in a single wake effect. The shadowing and no shadowing. The area of the turbine under
wind turbine T7 is operating in multiple wake effect because it shadowing can be calculated, provided that all the turbines in the
faces the wake from two upstream turbines T1 and T4. farm have the same diameter by ð2r o Þ as given in the following
The calculation of wind speed deficit at the position of T7 has equation:
been done by using Eq. (8): When a wind turbine faces multiple ! !
wake effect from upstream wind turbines, the resulting velocity vi   2 1 Lij  1 dijLij
Ashadow;i ¼ r i xij cos   2
þ r o cos  dij zij ð9Þ
can be calculated by equating the sum of the kinetic energy defi- r i xij r i xij
cits of each wake to the kinetic energy deficit of the mixed wake at
where xij is the distance between upstream wind turbine i and the
that point.
2 vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi3 downstream wind turbine j. r i ðxij Þ is the radius of the wake behind
u Nt  
uX v 2 the upstream wind turbine at distance xij and dij is the distance
vi ¼ vo 41  t 1 i 5 ð8Þ between the centers of downstream turbine and the center of the
i¼1
vo
wake effect, i.e. ðr i ðxij Þ þ r o Þ.

The multiple wake effects can be combined into a single wake 2.1.2. Larsen's model
effect in large wind farms where many turbines are placed toge- The model by G.C. Larsen, referred to as the Larsen's model, also
ther with a specific layout. This combined effect can be defined by known as the EWTSII model (European Wind Turbine Standards II)
using a mathematical formula to show the relations between these is based on the Prandtl turbulent boundary layer equations and
multiple wakes in a wind farm. There exist detailed models con- has closed form solutions for calculating the width of the wake
sidering the shadowed areas of the upstream wind turbines. This and the mean velocity profile in the wake. In order to obtain the
effect is called partial shadowing as shown in Fig. 6. The sha- closed form solutions, a self-similar velocity profile is assumed and
dowing is a measure of the degree of overlap between the area Prandtl's mixing length theory is used. The flow is further assumed
swept by the turbine experiencing shadowing Ao and the area to be incompressible and stationary, while the wind shear is
spanned by the wakes shadow cone Ashadow;i . neglected, hence the flow is axisymmetric. Larsen showed both
There are four distinct possible shadowing conditions, which a first order and a second order approximate solution to the
are complete shadowing, quasi complete shadowing, partial boundary layer equations [57].
 R. Shakoor et al. / Renewable and Sustainable Energy Reviews 58 (2016) 1048–1059 1053

Shown below are the governing equations of the first order 2.2.1. Eddy viscosity/2D field models
solution, in which the values for the rotor wake radius Rw and the Two dimensional field models assume axial symmetry in the
axial velocity deficit in the wake ΔU 1 are obtained. Where C T wakes. Since this leads to fewer equations to be solved simultaneously,
represents the thrust coefficient of wind turbine, V 1 is the free less time is needed for the calculation of the flow field. This approach
stream wind velocity and A is the rotor swept area of the wind was first used in [23] and several models related to this study. Ainslie
turbine. [23] was the first to model the effect of wake meandering on wake
deficits by relating wake meandering to the variability in wind
 15
15
35 1 direction. This model uses axis symmetrical and time averaged Navier
Rw ðxÞ ¼ 3C 21 ðC T Aðx ¼ xo ÞÞ3 ð10Þ
2π Stokes equations for incompressible flow with eddy viscosity (EV)
closure to calculate wake behavior. The EV model was implemented
"
V1 1
originally by Robinson and Neilson in the WINDOPS software, which
ðC T Aðx  xo Þ  2 Þ3 r 3 ð3c21 C T Aðx þ xo ÞÞ 2
1 2
ð∇U Þ1 ðx; r Þ ¼  became WindFarmer [62]. A similar implementation of the Eddy
9
Viscosity model was developed by Robinson in OpenWind [63]. The
 103 #2
35 Farm Layout Program (FLaP) [55] was developed by the University of
2 51
 ð3c1 Þ ð11Þ
2π Oldenburg (UO), also based on the Ainslie's model.
The EV model is used to solve an axisymmetric form of the
where c1 represents the non-dimensional mixing length and is Navier Stokes equations; it therefore qualifies as a simple RANS
given by Eq. (12), l ¼Prandtl's mixing length. model. The wake decay is dictated by the rate of mixing of
momentum from the surrounding flow into the wake zone, as
1
c1 ¼ lðC T AðxÞÞ3 ð12Þ determined by the ambient turbulence. With the construction of
offshore wind projects of significant size, it has become apparent
The parameter c1 is not effected by the change rotor size and that the standard Park and EV models tend to underestimate the
design. However, the wake effect is strongly depends upon the wake losses in offshore arrays [64]. This may be in part because
rotor dimensions. Gaumond [58] stated that for the row of wind the models assume that wind turbines have no effect on the pla-
turbines with narrow space 31 to 51 , the Larsen wake model netary boundary layer (PBL) other than the wakes they directly
underestimates the energy production. This is because of that the generate. The Eddy Viscosity model predicts direct wake drops,
uncertainty in the direction of wind is not modeled. especially the initial power drop, with the highest degree of
accuracy, but performs poorly under non-direct wake scenarios
2.1.3. Frandsen's model [50].
Frandsen introduced surface drag induced internal boundary
layer approach, which modifies the wind speed profile within the 2.2.2. Three dimensional field models
Planetary Boundary Layer (PBL) according to the increasing dis- The boundary layer models belong to Parabolic models. Ainslie
tance downstream in front of wind turbine [16]. The model dis- [23] applied a parabolic eddy viscosity model with steady flow and
tinguishes three different wake regimes: in the first regime, single axial symmetry, neglecting the pressure gradients in the outer flow
or multiple wake flow is present without interaction between [65]. In order to calculate the pressure gradient, a free vortex wake
neighboring wakes. The second regime starts when two neigh- model was used. Using this improved model similarly results in the
boring wake flows interact. The expansion of the wake is then wake farm with tuned parameters [27]. Energy Research Center
limited to an expansion in vertical direction only. The wake is in (ECN) has developed a wake farm model based on a modification of
the third and last regime, when the wake flow is in balance with the UPMWake model by Universidad Polytecnica de Madrid wake
the Planetary Boundary Layer [59]. This is indicated when the farm. It is essentially a 3D parabolized Navier Stokes code for the far
wind farm can be seen as infinitely large. Similar to the Jensen's wake using a k–ε turbulence model. This hat shape was changed
model, the velocity deficit in the wake is assumed to be hat into a Gaussian velocity deficit during the (Efficient Development of
shaped. Later studies showed that a different description of the Offshore WindFarms) ENDOW project [66].
wake diameter corresponded better with measurements in the far Crespo and Hernández [25] had drawn a parabolic code to a
wake. In the case of composite wakes (wakes consisting of mul- fully elliptic sort to solve divergences with experimental findings
tiple wakes originating from upstream turbines), the wakes are that are mainly seen in the near wake region, but no major mea-
divided in several sections (called mosaic tiles); each having a sures were taken in single wake calculations. Several elliptic field
constant but different velocity deficit. To calculate the mean wind models have been developed to study the flow around wind tur-
speed over the rotor area, a semi linear method is used. For more bines and through wind farms. In these models, wind turbines are
details, research paper [60] can be referred to regarding the modeled either using generalized actuator disks [67,68] or actua-
tor lines [69].
method for wake combination. Tong in reference [61] investigated
the performance of different analytical models for estimation of
power production and it is observed that for the single and mul-
3. Comparative study of different wake models
tiple wake effect test, the Frandsen wake model predict a larger
initial wake expansion and highest wind speed as compared to the
The overall wake aerodynamics of wind turbine has been
other analytical models [50]. Which shows that this model pro-
extensively discussed in previous reviews [46,47,70–74]. Among
vides additional power drop for the downstream turbines.
these, Vermeer et al. [46] conducted experiments ascertaining
mathematical models that precisely explain the effect of wake;
2.2. Field models mutually in terms of turbulence intensity and the decrease of wind
speed. From all models, some are only effective for near wake and
Field models are used to calculate the complete flow field others are only valid for far wake boundaries. Numerous studies
through a wind farm, or a part of the wind farm when the wind which performed broad evaluations between different wake
farm is regular, hence the name field models. In order to do so, models [19,75] allow us to conclude that there is a possibility of
field models function to solve the RANS (Reynolds-averaged high prediction accuracy error in all model performance as com-
Navier–Stokes equations) with a turbulence model for closure. pared to real measurements. It is because the downstream
1054 R. Shakoor et al. / Renewable and Sustainable Energy Reviews 58 (2016) 1048–1059

distance in the direction of wind highly influence on the wake Table 1 enlists the prediction error rates of the average wind
effect measurements of all wake models. A comparison of different speed and wake width by Eddy Viscosity, Frandsen, Larsen and
wake models presented in [48] does not suggest any particular Jensen's wake models. In the case of large downstream distance,
difference in terms of accuracy between the sophisticated and the Larsen and Jensen's wake models gives high prediction error,
simplified models. It is rather difficult to extricate one specific while the eddy viscosity model displays high prediction accuracy
model over the others in term of the calculation of wake effects for wake width measurements. In the case of less downstream
among the turbines. Even if more accurate computational distance, the Larsen and Frandsen's models show high prediction
approaches have been suggested [76]. accuracy, while the eddy viscosity and Jensen's wake models dis-
The most extensively used models are the analytical wake play some prediction error for average wind speed measurements.
models because of their minimal computational effort for multiple
wake simulation. Although the Jensen's model is one of the oldest
wake models, it is still very effective in wind farm layout optimi- 4. Wind farm layout optimization using Jensen's wake model
zation because of its simplicity in executing the optimization
procedures and simulation. The wind farm optimization can be executed by applying spe-
Vanluvanee [50] compared the practical and simulation results cific objective function. The most extensive method is by max-
using different wake models. Three different wake models were imizing the total produced power of wind farm. The power pro-
used for simulation i.e. Ainslie Eddy Viscosity, Jensen and Larsen duced power is contingent upon the total number of wind turbines
wake model. To assess the model performance, three parameters in a farm and their positions with respect to one another in order
were investigated: direct wake power drops, turbine direction to reduce the wake effect [78]. Wind farm layout optimization is
versus power and annual energy production. It was concluded that referred as the optimization task that chooses the best turbine
the Eddy Viscosity model worked betters in terms of direct wake positions. Optimization does not necessarily mean finding the
power drop but dropped its performance within the non-direct optimum solution to a problem, since it may be unfeasible due to
wake power drop. In general, the performance of Jensen's wake the characteristics of the problem, which in many cases are
model was found to be the best following the Eddy Viscosity model. included in the category of NP-hard problems [79]. In literature, it
F. Seim [75] validated three kinematic wake models using can be found that there is a trade-off between the layout of the
WindSim software in complex terrain condition for eight single- wind farm and the energy production, as some papers have shown
wake cases. He concluded that for all cases, the Larsen's model that modeling the non-uniform layout could produce more energy
overestimated the width of wake, however it gave a constant output as compared to uniform layouts [29,80–85]. On the other
offset which reduced the uncertainty in the power loss calculation. hand, this type of papers clearly indicates that the researchers are
S. Jeon et al. [77] also reached the same results by applying keen to get maximum energy from the wind farm and not inter-
these models at a commercially operated onshore wind farm. He ested on the turbine's shape. Currently, there are several com-
found that Jensen's wake model outperformed in comparison with mercial programs that enable the wind resource to be assessed at
other wake models. However, Eddy viscosity and Larsen models the placement. Reference [86] summarizes these software and
predicted the width of the wake with relatively high degree of their difference.
accuracy. There are a number of optimization techniques which have
Moreover, EMD International of Denmark accomplished a study successfully been used in solving wind farm layout problems, such
at three different offshore wind farms in Denmark. They per- as Genetic Algorithms [35], Simulated Annealing [37,38], Differ-
formed the simulation based results using WindPRO to calculate ential Evolution (DE) [39], Simulated Evolution (SimE) [40,41], Ant
the wind farm efficiency estimated by Jensen, Larsen and Eddy Colony Optimization (ACO) [42], Particle Swarm Optimization
viscosity models. The study presented the comparison of simula- (PSO) [43], Stochastic Evolution [44,45], Definite Point Selection
tion results with the practical value of efficiency measured at these (DPS) [87], Bionic Optimization, Gradient based optimization,
wind farms [76]. Compared with the other two models that Numerical added simulation and Monte carlo optimization
underestimated the array losses in the wind farms, the results technique.
showed that the park efficiency value predicted by the Jensen The next few paragraphs discuss some significant published
model was the closest to the actual measurement value. works related to the topic. In this section, we analyze critically the
Furthermore, Grumond et al. [58] validated the performance of points that the researches had identified during study and review
different wake models at Lillgrund (Sweden) and Horns Rev their results.
(Denmark) wind farms. They concluded that due to the reason of Mosetti et al. [84] was the first attempt to use Jensen's wake
small spacing between the wind turbines in Lillgrund wind farm, model with genetic algorithm in solving wind farm optimization
all wake models gave large prediction error. The Frandsen's and problem. The objective of the research was to get the best location
Jensen's wake models required a thrust coefficient of wind turbine of each turbine in wind farm. This resulted in the reduction of cost
and freestream wind speed as an input parameters, while the and an increase in energy output. They conducted the research
Larsen and Eddy Viscosity needed the value of ambient turbulence by describing two fitness functions, one for the reduction of cost,
intensity additionally as an input parameter. and the other for maximizing the energy. Then, they formulated a

Table 1
Error rate of wake models for average wind speed and wake width prediction at different downstream distance [77].

Downstream distance Frandsen's model error rate (%) Jensen’s model error rate (%) Larsen’s model error rate (%) Eddy viscosity model error rate (%)

Wake width Average wind Wake width Average wind Wake width Average wind Wake width Average wind speed
speed speed speed

2.55D 10 6 30 14  10 6  20 5
3.75D 23 3 38 12 0 3  15 4
5.1D 45 2 55 12 30 2 10 6
7.3D 63 0 71 7 54 1 20 1
 R. Shakoor et al. / Renewable and Sustainable Energy Reviews 58 (2016) 1048–1059 1055

single equation and aggregated the two fitness functions. Each Youjie et al. [89] utilized the Jensen's wake model for wake
objective function was consigned the weights depending on the deficit calculations and for various wind conditions. They per-
preferences given to each optimization factor. They used the 100 formed the modeling of wind farm to achieve the exact wind
square grid farm layout uniformly and the centers of each square speed at placement position of wind turbine, and at the same time
grid were likely to be location of the wind turbine. They specified obtained the total power output of the wind farm. They used
the following parameters for the wind farm modeling; diameter quadratic interpolation optimization method for wind farm opti-
D ¼40 m, hub height Z¼ 60 m and a constant thrust coefficient mization. Eq. (17) gives the objective function of Youjie et al. [89].
CT ¼0.88. The cell size was equal to 5D and the wind turbine was 1
placed at the midpoint of the cell. They described the optimization min f ðvÞ ¼  ρAr C p ðvi Þ3 ð17Þ
2
function mathematically as:
In the above study, 120 wind generators were used which
1 ct produced 240 MW. The wind farm contained twenty equally
Objective A ¼ x1 þ x2 ð13Þ
pt pt spaced column and six rows. In the parallel wind speed direction,
the distance between two adjacent wind turbines was set as
where pt is the total produced power in one year, x1 and x2 are
300 m. The hub height was the same for all the turbines. Their
randomly chosen weights, and ct is the cost per annum of the wind
simulation results showed that mean output power without con-
farm. They analyzed the problem by changing the wind scenarios
sidering the wake effect was almost 0.7 MW. On the other hand,
i.e. constant wind speed with uniform direction, constant wind
the mean power output became 0.5 MW while taking into account
speed with variable direction, variable wind speed with variable
the wake effect. Therefore, the wake effect reduced the output
direction.
power by 28%. It was also concluded that wake phenomenon
In all three wind scenarios, Mosetti et al. [84] installed a very
affected the power output to a large extent. However, a very
limited number of wind turbines which may improve the overall simply analysis for the power output with and without wake effect
efficiency of wind farm performance due to minimum wake has been done, and the investigation by changing the layout of
interactions. It is obvious that the small number of wind turbines wind farm on its performance was not implemented.
gives high efficiency but it is actually the waste of wind farm area Herbert Acero [90] studied the systematic analysis of the
resource which will become the reason to increase the objective optimal positioning of wind turbine in a one-dimensional
function of the wind farm optimization. arrangement, lined-up in the direction of the wind by consider-
No work has been done after Mosetti et al. [84] within the ing the wake effect. The optimal positioning of wind turbines in a
duration of 11 years. Then the wake decay effect in wind farm straight line, on flat terrain, by considering wake effects had been
design was studied by employing Jensen's wake model by Grady conducted using both simulated annealing (SA) and genetic algo-
et al. [81]. They replicated the experiments presented in [84] by rithms (GA). The main emphasis of the study was to focus on the
modifying the settings of the GA. They employed more individuals power produced by a group of turbines as described by the fol-
(600) and generations (3000) in GA to achieve better layout for the lowing equation:
wind farm. They modeled the wind farm of 50D  50D size, dis-
1
tributing the wind turbines in squares having equal length and minimize pðU o Þ ¼ ρAC p ðU o ÞU 3o ð18Þ
2
width. Each cell had a width of 5D. The optimization was done
using the fitness function given by Eq. (14), minimizing the cost where Uo is the incoming free-flow wind speed, A is the area
per unit power produced; swept by the rotor and ρ is the air density, and C p ðU o ) is the power
coefficient of the turbine. Two test scenarios were considered by
cost
Minimize ð14Þ employing the following procedure: In the first scenario, constant
ptot wind turbine hub height and unidirectional wind flow were
where Ptot is the power taking out from N turbines, and the cost is assumed, whereas, in the second scenario, they assumed bidirec-
expressed by Eq. (15): tional wind flow.
  Emami and Noghreh [91] conducted quite a different study
2 1 2
cost ¼ N þ e  0:00174N ð15Þ from the previous author's works in terms of algorithm and
3 3 objective function. They made novel changes in the GA algorithm
Mora et al. [88] used Jensen's far wake model and suggested GA and compared the obtained results with Mosetti et al. [84] and
for the optimal layout of wind farm. They optimized the wind farm Grady et al. [81]. He showed that the developed objective function
layout problem in term of Net Present Cost (NPC) by taking into is more accurate in terms of efficiency, output power and cost in
account the single objective function represented by the following comparison with those presented by [81,84]. Mathematically, the
equation: fitness function was defined by using the multiobjective optimi-
zation function as below:
N 1 ðxÞ Nt ðxÞ
NPC ðx; i; t Þ ¼ þ…þ  ICðxÞ ð16Þ 1
1þi ð1 þ iÞt g ¼ ω1 costm þ ω2 ð19Þ
ptotal
where IC is the initial capital cost, i is the rate of discount on initial
capital, x represents the state vector with the height and location ω1 þ ω2 ¼ 1 ð20Þ
of wind turbines, and t is the life time of the project. In this study, where ptotal abbreviates the energy generated in one year (MW/yr),
the cell size considered was 20  20 (arbitrary length units). This ω1 and ω2 are capricious preferred weights. Cost represents the
research focused on three test cases. The direction and speed of per unit value of cost/year of wind farm. To calculate the wake
the wind were set as constant in the first two cases. They calcu- deficit Jensen's model was used. The wind scenario and wind farm
lated the NPC for the hilly areas where the wind speed was varying cell parameters were considered similar as used by [81,84]. The
in the third test case. They proposed GA for all three cases and location of the wind turbine was considered in a discrete square
showed that cost optimization can be done in optimal way. shaped cell. For the first wind scenario, the proposed method
However, they did not compare their results with any other related reduced the number of generations and chromosomes in com-
research work. Also, the total area of the wind farm and its specific parisons to the Mosetti et al. [84] while obtaining the same results.
shape are not given. For the second and third wind scenarios, the improved results
1056 R. Shakoor et al. / Renewable and Sustainable Energy Reviews 58 (2016) 1048–1059

were attained. They consider homogeneous wind turbine and as follows:

constant area of wind farm. However, the variable height and rotor
X
Nt
diameter of wind turbines also affect the total output of the MAX pi ð22Þ
wind farm. i¼1

Another researcher [92] employed the same fitness function as where pi is the total power of ith turbine. It was concluded that the
used by [81,84]. He contributed in term of reduction of size of the use of ACO algorithm can help to find better wind farm layouts
wind farm by reducing the spacing. He optimized wind farm lay- compared to that in prior studies without being trapped in local
out by using GA for discrete solution space. He simulated the maximum in selected problem within a reasonable solution time.
results for same three scenarios as proposed in Mosetti. The The performance of the proposed algorithm was generally better
reduction of grid spacing resulted in the decrease in cost per unit than that of existing algorithms proposed for continuous pro-
power for all cases. Mittal [92] recommended the use of micro blems; thus, it was obvious that the algorithm by using ACO was
siting technique with genetic algorithm (GA) in order to find more useful for finding global maximum such as their continuous
precise locations of turbines inside a wind farm. He proposed the function.
cell size of 1 m  1 m and obtained the reduced cost per unit A new objective named the Turbine-Site Matching Index (TSMI)
power in comparison with the result from the study by Grady et al. was introduced by [97]. They analyzed the layout problem using
[81] and Mosetti et al. [84] for all three scenarios. The obtained greedy algorithm in which the capital cost and the capacity factor
results revealed that the modification in grid spacing reduced cost (CF) were analyzed with respect to the change in the hub height of
per unit power in the range of 11–16% for the three test scenarios. wind turbines. The objective function used in this study is given by
Rahmani et al. [93] was the first to utilize PSO for solving the the Eq. (23).
wind farm layout design problem in order to increase the wind ptot
NP rated
farm output power. They obtained the same objective function as object ¼ TSMI þ ð23Þ
ICC
used by [81] and simulated the program by changing the number
where ptot is total power output, P rated is rated power output, N is
of wind turbines. The simulation results exposed an aptness of the
number of wind turbines and ICC is the normalized initial capacity
suggested PSO to solve the wind farm layout problem.
cost. They divided their research into two portions. In the first
Moreno et al. [94] used seeded GA to optimize the cost of the
part, flexible power curve model was developed with no power
wind farm using Jensen's wake model. They introduced an oro-
control criteria. They assumed the uniform wind speed and the
graphy model and the shape model for the first time. The novelty uniform wind direction for this case. Three conditions are studied
of the work was to use the seed for the proposed GA. The fitness in this research. Condition 1: The tower height had no impact on
function was given by the following equation: the wind farm wake model and the distance among the wind
turbines. Condition 2: The relative spacing between the wind
X
n X
φ ¼ Bt  N:C i  C cij ð21Þ turbines are fixed and the wake model changes with the tower
i ¼ 0 jo1 height. Condition 3: As the tower height changes, the wake model
and the distance among the wind turbines varies. In second por-
where C i is the installation cost of wind turbines, C cij is the cost of tion, they investigated the power curve model with power control
connection between road construction and wind turbines which mechanisms. For this, it is assumed that the turbine wake model
depends on the location of the specific turbine, Bt depends on the and the distance amongst the wind turbines depend on the tower
payback period of the wind farm, t is the life of project and N height. In this study, they used 40 m diameter wind turbines with
denotes the total number of wind turbines in the wind farm under hub height varies from 40 m to 600 m. The effectiveness and the
consideration. The authors of this paper used seeded GA with applicability of proposed method is illustrated through the
greedy heuristics for population initializing. Additionally to com- numerical studies. They concluded that the tower height had not
plete the GA's initial population, some individuals were included much impact on the optimization of wind farm layout.
and created randomly. An integer matrix was used for the opti- Reference [98] investigated the effects of varying hub heights of
mization of the problem. They considered 15 different sets of wind wind turbines within wind farm on the power output. The authors
speed for different orography of wind farm, and based on these finally reached the result that the power output became better in
sets, random shapes of wind farm were generated. Then, a pro- case of changing the hub height of different wind turbines from
posed GA was applied on them. The parameters used in this 50 m to 78 m, even when using the same number of wind tur-
bines. They also analyzed different cost models and gave con-
research were, wind farm cell size of 10 m  10 m and the dia-
cluding remark that one can reduce the cost per unit power pro-
meter of wind turbine 90 m. The comparison between different
duced by using different hub height wind turbines in the wind
approaches showed that the seeded GA was the best choice to be
farm. On the other hand, this is generally valid only for the large
used, followed by the unseeded GA for all test cases.
wind farm and large hub height variation.
Samina et al. [95] worked on the hypothetical layout of wind
Shakoor et al. [99] introduced a new concept, the effect of wind
turbine in a wind farm. They generated a Matlab coding using GA
farm area dimensions on the layout optimization. They used GA
solver with the same objective function as used by [81,84,92].
for the optimization of cost per unit power obtained from the
They worked on a very simple wind scenario i.e. constant wind wind farm. The outcome shows that area shape strongly affects the
with uniform direction and compared their results with the results total free stream wind velocity entering in the boundaries of wind
from previous studies. farm and wind turbine power output.
Yunus et al. [96] used Ant Colony Optimization (ACO) algorithm A binary particle swarm optimization (BPSO) was applied to
in order to solve the wind farm layout problem. A new mathe- locate the optimal position of the turbines within the wind farm
matical modeling was proposed assuming three different scenar- by [100]. They introduced the variable 'time varying acceleration
ios of the wind direction. They used the Cartesian coordinates coefficient (TVAC)' in their study. The BPSO–TVAC algorithm was
system for the solution of wind farm layout in order to calculate applied to a 10 m  10 m square shaped farm site, considering
the distance between the turbines. They assumed the circular uniform wind and non-uniform wind speed with variable direc-
shaped wind farm for the analysis and used the objective function tion characteristics. The analysis of cost versus maximum output
 R. Shakoor et al. / Renewable and Sustainable Energy Reviews 58 (2016) 1048–1059 1057

power of the wind farm was then performed. The parameters used modeling turbine locations i.e. continuous and discrete. A con-
were the same as in the previously discussed studies [81,84]. The tinuous location model allows turbines to be placed anywhere
objective function for the optimization layout was defined as: within the farm, subject to boundary and proximity constraints. A

 discrete location model only allows turbines to be placed at a finite
2 1  0:00174N 2
costðN Þ N 3 þ 3e number of places.
Fitness Function ¼ ¼ P360 PN ð24Þ
pT K¼0 i ¼ 1 f k pi ðui

where N is equal to the number of wind turbines installed in a

5. Conclusion
wind farm ð0 o N r 100Þ; ui is mean wind velocity depending on
P
wind distribution, f k is wind frequent distribution ð 360
k ¼ 0 f k ¼ 1Þ. It This paper focuses on the existing literature on the wake effect
is established that algorithm efficiency was maximum for smaller
modeling and explains the important distinction between far and
wind farms. On the other hand, for larger wind farms, the algo-
near wake effect in large wind farms. The detailed comparison of
rithm might not work much or end up with the same layout as in
wind turbine wake effect models in terms of wake effect predic-
previous studies.
tion and wind turbine power calculation is presented. The results
Turner et al. [101] developed mixed integer linear and quad-
have shown that the prediction accuracy of all wake models is
ratic optimization formulations and applied them to several
highly affected by the spacing and downstream distance between
example layout cases in the literature. Compared to previous
the wind turbines.
approaches, their models produced layouts that tended to be more
It is particularly difficult to predict the accuracy of the specific
symmetric and that generated slightly more power. They con-
models or group of models in term of multiple wake. The cap-
sidered a 10  10 grid, varying the number of turbines placed and
ability of wake models to predict atmospheric and sea stability
wind directions. The formulations were modeled in C þ þ and
effects and losses due to nearby farm may highly appear to be
solved using CPLEX 12.1. The fitness function is given by following
lacking. Increased spacing can undeniably decrease wake losses,
equation.
but wake models must be improved for the optimization of wind
X
N
X  farm layout. It is found that most related studies on wind farms
2
minimize yj diji yi ð25Þ utilize the Jensen's wake model which is the simplest available
j¼1
method to date. This model gives an acceptable degree of pre-
where dij is the wind speed deficit generated at position j due to a diction accuracy for the off-shore as well as for the flat terrain
turbine placed at position i, yj is the distance (along the direction onshore wind farms.
of the wind) between positions i and j, binary variables yi , define y The wind farm layout design is categorized as an NP-hard
to be the feasible region of formulation. In the study, wind turbine optimization problem containing large number of constraints.
with rotor diameter of 40 m and hub height of 60 m and thrust Majority of researchers are working to solve the problem by
coefficient of 0.88 was used. implementing the existing optimization techniques. On the other
Table 2 presents the work done on wind farm layout optimi- hand, only some researchers focus on the solution method itself.
zation by using Jensen's wake model for wake deficit measure- The main studies that have been carried out to date are certainly a
ment and different optimization methods for wind turbine posi- good starting point for further research to obtain more effective
tioning. It is clear from the table that most of the research has been solution methods, but they cannot be considered entirely satis-
done by using heuristic methods. However, it can also be done factory for several reasons. First, most of the studies consider a
by mathematical techniques. There are two main approaches for discrete domain dividing the total wind farm area space into small

Table 2
Different characteristics of wind farm optimization problems using Jensen's wake model, HH ¼hub height and RD ¼rotor diameter.

Ref, year Optimization method Equation no (Objective function) Wind behavior HH, RD (m) Computational domain

[84], 1994 GA 13 Mean 60,40 Discrete

[81], 2005 GA 14 Mean 60,40 Discrete
[88], 2007 GA 16 Mean Variable Discrete
[89], 2009 QIO 17 – Unspecified Discrete
[90], 2009 SA,GA 18 Mean Variable,77 Unspecified
[91],2010 GA 19 Weibull 60,40 Discrete
[92], 2010 GA 14 Mean Unspecified Discrete
[93], 2010 PSO 14 Mean 60,40 Continuous
[94], 2011 GA 21 Weibull Unspecified Discrete
[95], 2012 SS, GA 14 Mean 60,40 Discrete
[96], 2012 ACSA 22 Mean Unspecified Continuous
[97], 2013 GGA 23 Mean Variable,40 Continuous
[100], 2013 BPSO 24 Mean 60,40 Continuous
[101], 2014 MILQO 25 Mean 60,40 Continuous
[99], 2015 GA 14 Mean 60,40 Discrete
[87], 2015 DPS, GA 14 Mean 60,40 Discrete

GA – Genetic Algorithm.
QI – Quadratic Interpolation Optimization.
SA – Simulated Annealing.
PSO – Particle Swarm Optimization.
SS – Spread Sheet.
ACSA – Ant Colony Search Algorithm.
GGA – Global Greedy Algorithm.
BPSO – Binary Particle Swarm Optimization.
MILQO – Mixed Integer Linear and Quadratic Optimization.
DSP – Definite Point Selection.
1058 R. Shakoor et al. / Renewable and Sustainable Energy Reviews 58 (2016) 1048–1059

cells for wind turbine positioning. However, the continuous space [29] Aytun Ozturk U, Norman BA. Heuristic methods for wind energy conversion
search gives more optimal points for turbine installation. Second, system positioning. Electric Power Syst Res 2004;70:179–85.
[30] Youcef Ettoumi F, Adane AEH, Benzaoui ML, Bouzergui N. Comparative
for on-shore wind farm, the topography of the area must be simulation of wind park design and siting in Algeria. Renew Energy
described as one of the optimization objectives or constraints in 2008;33:2333–8.
layout optimization techniques. Third, except the relative positions [31] Mustakerov I, Borissova D. Wind turbines type and number choice using
combinatorial optimization. Renew Energy 2010;35:1887–94.
of the wind turbines in the installation site, the boundaries of the [32] Samorani M. The wind farm layout optimization problem. In: Handbook of
installation area also affect the final electricity production, so it wind power systems. Springer; 2013. p. 1–18.
must be considered as an optimization step. Fourth, the review of [33] Horowitz E, Sahni S. Fundamentals of computer algorithms. WH Freeman &
Co; 1978.
previous work also shows that most of the research has been done [34] Archer R, Nates G, Donovan S, Waterer H. Wind turbine interference in a
using heuristic techniques and even the share of genetic algo- wind farm layout optimization mixed integer linear programming model.
rithms is more than 75% for wind farm layout optimization. Hence, Wind Eng 2011;35:165–75.
[35] Goldberg DE, Holland JH. Genetic algorithms and machine learning. Mach
it is required to explore new optimization techniques. Further, Learn 1988;3:95–9.
different optimization approaches can be hybridized for solving [36] Holland H John. Adaptation in natural and artificial systems. Cambridge, MA:
the problem to get optimum results. MIT Press; 1992.
[37] Kirkpatrick S, Vecchi M. Optimization by simmulated annealing. Science
1983;220:671–80.
[38] Nahar S, Sahni S, Shragowitz E. Simulated annealing and combinatorial
optimization. In: Proceedings of the 23rd ACM/IEEE design automation
References conference; 1986. p. 293–9.
[39] Storn R, Price K. Differential evolution–a simple and efficient heuristic for
[1] Raheem A, Hassan MY, Shakoor R, Rasheed N. Economic feasibility of stand- global optimization over continuous spaces. J Global Optim 1997;11:341–59.
alone wind energy hybrid with bioenergy from anaerobic digestion for [40] King RM, Banerjee P. Optimization by simulated evolution with applications
electrification of remote area of Pakistan. Int J Integr Eng 2014;6:1–8. to standard cell placement. In: Proceedings of the 27th ACM/IEEE design
[2] Khan SA, Rehman S. Iterative non-deterministic algorithms in on-shore wind automation conference; 1990, p. 20–5.
farm design: a brief survey. Renew Sustain Energy Rev 2013;19:370–84. [41] Kling RM, Banerjee P. Esp: placement by simulated evolution. IEEE Trans
[3] Perveen R, Kishor N, Mohanty SR. Off-shore wind farm development: present Comput-Aided Design Integr Circuits Syst 1989;8:245–56.
status and challenges. Renew Sustain Energy Rev 2014;29:780–92. [42] Colorni A, Dorigo M, Maniezzo V. Distributed optimization by ant colonies.
[4] I Energy Agency. International Energy Agency WIND 2012 Annual Report In: Proceedings of the first European conference on artificial life; 1991. p.
{@ONLINE}. ed; 2012. 134–42.
[5] Fried L, Shukla S, Sawyer S. Global wind report: annual market update 2012. [43] Kennedy J, Eberhart R. Particle swarm optimization. In: Proceedings of
GWEC. 〈http://www.gwec.net/publications/globalwind-report-2〉; 2012. ICNN'95 – international conference on neural networks, vol. 4; 1995, p.
[6] Kinab E, Elkhoury M. Renewable energy use in Lebanon: barriers and solu- 1942–8.
tions. Renew Sustain Energy Rev 2012;16:4422–31. [44] Saab Y, Rao V. An evolution-based approach to partitioning ASIC systems. In:
[7] Saidur R, Islam MR, Rahim NA, Solangi KH. A review on global wind energy Proceedings of the 26th ACM/IEEE design automation conference; 1989. p.
policy. Renew Sustain Energy Rev 2010;14:1744–62. 767–70.
[8] Sahu BK, Hiloidhari M, Baruah DC. Global trend in wind power with special [45] Saab YG, Rao VB. Combinatorial optimization by stochastic evolution. IEEE
focus on the top five wind power producing countries. Renew Sustain Energy Trans Comp-Aided Design 1991;10:525–35.
Rev 2013;19:348–59. [46] Vermeer LJ, Sørensen JN, Crespo A. Wind turbine wake aerodynamics. Prog
[9] Cheng S, Li Z, Mang H-P, Huba E-M. A review of prefabricated biogas Aerosp Sci 2003;39:467–510.
digesters in China. Renew Sustain Energy Rev 2013;28:738–48. [47] Sanderse B. Aerodynamics of wind turbine wakes literature review. Energy
[10] Stambouli AB, Khiat Z, Flazi S, Kitamura Y. A review on the renewable energy Research Centre of Netherlands ECN-E-09-016; 2009.
development in Algeria: current perspective, energy scenario and sustain- [48] Renkema DJ. Validation of wind turbine wake models (Master of Science
ability issues. Renew Sustain Energy Rev 2012;16:4445–60. Thesis). TUDelft; 2007.
[11] Abanda FH. Renewable energy sources in Cameroon: potentials, benefits and [49] Frandsen S, Thogersen ML. Integrated fatigue loading for wind turbines in
enabling environment. Renew Sustain Energy Rev 2012;16:4557–62. wind farms by combining ambient turbulence and wakes. Wind Eng
[12] Komarov D, Stupar S, Simonović A, Stanojević M. Prospects of wind energy 1999;23:327–40.
sector development in Serbia with relevant regulatory framework overview. [50] VanLuvanee DR. Investigation of observed and modelled wake effects at
Renew Sustain Energy Rev 2012;16:2618–30. Horns Rev using WindPRO. Technical University of Denmark Department of
[13] Jervase JA, Al-Lawati AM. Wind energy potential assessment for the Sulta- Mechanical Engineering; 2006.
nate of Oman. Renew Sustain Energy Rev 2012;16:1496–507. [51] González-Longatt F, Wall P, Terzija V. Wake effect in wind farm performance:
[14] Moskalenko N, Rudion K. Study of wake effects for offshore wind farm steady-state and dynamic behavior. Renew Energy 2012;39:329–38.
planning; 2010. [52] Han T. The assessment of dynamic wake effects on loading (MS Thesis). The
[15] Jensen NO. A note on wind generator interaction. Technical report Riso-M- Netherland: Department of Aerospace Engineering Delft University of
2411; 1983. Technology; 2011.
[16] Frandsen S, Barthelmie R, Pryor S, Rathmann O, Larsen S. Deficit in large [53] Barthelmie RJ, Larsen GC, Frandsen ST, Folkerts L, Rados K, Pryor SC, et al.
offshore wind farms; 2006. p. 39–53. Comparison of wake model simulations with offshore wind turbine wake
[17] Katic I, Højstrup J, Jensen N. A simple model for cluster efficiency. In: Eur- profiles measured by sodar. J Atmos Oceanic Technol 2006;23:888–901.
opean Wind Energy Association Conference and Exhibition; 1986. p. 407–10. [54] Barthelmie RJ, Rathmann O, Frandsen ST, Hansen KS, Politis E, Prospatho-
[18] Ishihara T, Yamaguchi A, Fujino Y. Development of a new wake model based poulos J, et al. Modelling and measurements of wakes in large wind farms. J
on a wind tunnel experiment. Global Wind Power 2004 p. 1. Phys: Conf Ser 2007;75:012049.
[19] Werle M. A new analytical model for wind turbine wakes. Report no. FD. Vol. [55] Lange B, Waldl H-P, Guerrero AG, Heinemann D, Barthelmie RJ. Modelling of
200801; 2008. offshore wind turbine wakes with the wind farm program FLaP. Wind
[20] Porté-Agel F, Bastankhah M. A new analytical model for wind-turbine wakes; Energy 2003;6:87–104.
2014. [56] Zhang X, Wang W. Wind farm and wake effect modeling for simulation of a
[21] Manwell JF, McGowan JG, Rogers AL. Front matter and index. Wiley online studied power system. In: Power systems conference and exposition.
library; 2002. PSCE'09. IEEE/PES; 2009, p. 1–6.
[22] Lissaman PS. Energy effectiveness of arbitrary arrays of wind turbines. J [57] Larsen GC. A simple wake calculation procedure. Rosklide, Denmark: Risø
Energy 1979;3:323–8. National Laboratory; 1988 Technical report Risø-M-2760.
[23] Ainslie JF. Calculating the flowfield in the wake of wind turbines. J Wind Eng [58] Gaumond M, Réthoré PE, Bechmann A, Ott S, Larsen GC, Pena Diaz A et al.
Ind Aerodyn 1988;27:213–24. Benchmarking of wind turbine wake models in large offshore windfarms. In:
[24] Crespo A, Hernández J. Numerical modelling of the flow field in a wind Proceedings of the science of making torque from wind; 2012.
turbine wake. In: Proceedings of the 3rd Joint ASCE/ASME Mechanics Con- [59] Frandsen ST. Turbulence and turbulence-generated structural loading in
ference; 1989. p. 121–7. wind turbine clusters. Roskilde, Denmark: Risø National Laboratory; 2007.
[25] Crespo A, Hernández J. Parabolic and elliptic models of wind-turbine wakes, [60] Rathmann O, Frandsen S, Barthelmie R. BL3. 199 wake modelling for inter-
application to the interaction between different wakes and turbines. Phoe- mediate and large wind farms; 2007. p. 1–8.
nics J Comput Fluid Dyn Appl 1991;4:104–27. [61] Tong W, Chowdhury S, Zhang J, Messac A. Impact of different wake models
[26] Schepers J, Van der Pijl S. Improved modelling of wake aerodynamics and on the estimation of wind farm power generation. In: Proceedings of the
assessment of new farm control strategies. J Phys: Conf Ser 2007;75:012039. 12th AIAA aviation technology, integration, and operations (ATIO) con-
[27] Van der Pijl S, Schepers J. Improvements of the WAKEFARM wake model. ference and 14th AIAA/ISSMO multidisciplinary analysis and optimization
Energy Research Center of Netherland 2006:2. conference, American Institute of Aeronautics and Astronautics; 2012.
[28] Crasto G, Gravdahl AR. CFD wake modeling using a porous disc. In: European [62] WindFarmer G. Wind farm design software: theory manual. Garrad Hassan
Wind Energy Conference Exhibition; 2008. and Partners Ltd; 2010.
 R. Shakoor et al. / Renewable and Sustainable Energy Reviews 58 (2016) 1048–1059 1059

[63] Brower M, Robinson N. The open wind deep-array wake model: develop- [85] Şişbot S, Turgut Ö, Tunç M, Çamdalı Ü. Optimal positioning of wind turbines
ment and validation. USA: AWS Truepower; 2010. on Gökçeada using multi-objective genetic algorithm. Wind Energy
[64] Schlez W, Neubert A, Hassan G, Gmbh D, Str MC. New developments in large 2010;13:297–306.
wind farm modelling. [86] Serrano González J, Burgos Payán M, Santos JMR, González-Longatt F. A
[65] Riedel V, Neumann T, Strack M. Beyond the ainslie model: 3D Navier Stokes review and recent developments in the optimal wind-turbine micro-siting
Simulation of wind flow through large offshore wind farms; 2006. p. 3–6. problem. Renew Sustain Energy Rev 2014;30:133–44.
[66] Schepers JG. ENDOW: validation and improvement of ECN's wake model; [87] Shakoor R, Hassan MY, Raheem A, Rasheed N, Mohd Nasir MN. Wind farm
2003. layout optimization by using definite point selection and genetic algorithm.
[67] Ammara I, Leclerc C, Masson C. A viscous three-dimensional differential/ In: 2014 IEEE international conference on power and energy (PECon 2014);
actuator-disk method for the aerodynamic analysis of wind farms. J Sol 2014. p. 191–5.
Energy Eng 2002;124:345–56. [88] Castro Mora J, Calero Barón JM, Riquelme Santos JM, Payán M Burgos. An
[68] Mikkelsen R. Actuator disc methods applied to wind turbines. Kongens evolutive algorithm for wind farm optimal design. Neurocomputing
Lyngby, Denmark: Technical University of Denmark; 2003. 2007;70:2651–8.
[69] Troldborg N, Sørensen JN, Mikkelsen RF. Actuator line modeling of wind [89] Ma Y, Yang H, Zhou X, Li J, Wen H. The dynamic modeling of wind farms
turbine wakes. Kongens Lyngby, Denmark: Technical University of Denmark, considering wake effects and its optimal distribution. In: Proceedings of the
Danmarks Tekniske Universitet, Department of Energy Engineering Institut 2009 world non-grid-connected wind power and energy conference; 2009.
for Energiteknik; 2008. p. 1–4.
[70] Snel H. Review of the present status of rotor aerodynamics. Wind Energy [90] Herbert-Acero JF, Franco-Acevedo JR, Valenzuela-Rendón M, Probst-
1998;1:46–69. Oleszewski O. Linear wind farm layout optimization through computa-
[71] Hansen MOL, Sørensen JN, Voutsinas S, Sørensen N, Madsen HA. State of the tional intelligence. In: MICAI 2009: advances in artificial intelligence. Berlin,
art in wind turbine aerodynamics and aeroelasticity. Prog Aerosp Sci Heidelberg: Springer; 2009. p. 692–703.
2006;42:285–330. [91] Emami A, Noghreh P. New approach on optimization in placement of wind
[72] Sanderse B, Pijl SP, Koren B. Review of computational fluid dynamics for turbines within wind farm by genetic algorithms. Renew Energy
wind turbine wake aerodynamics. Wind Energy 2011;14:799–819. 2010;35:1559–64.
[73] Crespo A, Hernandez J, Frandsen S. Survey of modelling methods for wind [92] Mittal A. Optimization of the layout of large wind farms using genetic
turbine wakes and wind farms. Wind energy 1999;2:1–24. algorithm (MS thesis). Cleveland, Ohio: Department of Mechanical and
[74] Snel H. Review of aerodynamics for wind turbines. Wind Energy Aerospace Engineering Case Western Reserve University; 2010.
2003;6:203–11. [93] Rahmani R, Khairuddin A, Cherati Sm, Mahmoud Pesaran HA. A novel
[75] Seim F. Validating kinematic wake models in complex terrain using CFD method for optimal placing wind turbines in a wind farm using particle
(Master thesis). Akershus, Norway: Department of Ecology and Natural swarm optimization (PSO). In: 2010 Conference Proceedings IPEC; 2010. p.
Resource Management, Norwegian University of Life Sciences; 2015. 134–9.
[76] Sørensen T, Thøgersen ML, Nielsen P, Jernesvej N. Adapting and calibration of [94] Saavedra-Moreno B, Salcedo-Sanz S, Paniagua-Tineo a, Prieto L, Portilla-
existing wake models to meet the conditions inside offshore wind farms. Figueras a. Seeding evolutionary algorithms with heuristics for optimal wind
Aalborg: EMD International A/S; 2008. turbines positioning in wind farms. Renew Energy 2011;36:2838–44.
[77] Jeon BK Sanghyeon, Huh Jongchul. Comparison and verification of wake [95] Rajper S, Amin IJ. Optimization of wind turbine micrositing: a comparative
models in an onshore wind farm considering single wake condition of the study. Renew Sustain Energy Rev 2012;16:5485–92.
2 MW wind turbine. Energy 2015:1769–e1777. [96] Eroğlu Y, Seçkiner SU. Design of wind farm layout using ant colony algo-
[78] Samorani M. The wind farm layout optimization problem. In: Handbook of rithm. Renew Energy 2012;44:53–62.
wind power systems. Berlin, Heidelberg: Springer; 2013. p. 21–38. [97] Chen K, Song MX, Zhang X. The investigation of tower height matching
[79] Gary M, Johnson D. Computers and intractability a guide to the theory of NP- optimization for wind turbine positioning in the wind farm. J Wind Eng Ind
completeness. San Francisco: WH Freman and Co; 1979. Aerodyn 2013;114:83–95.
[80] Rivas RA, Clausen J, Hansen KS, Jensen LE. Solving the turbine positioning [98] Chen Y, Li H, Jin K, Song Q. Wind farm layout optimization using genetic
problem for large offshore wind farms by simulated annealing. Wind Eng algorithm with different hub height wind turbines. Energy Convers Manag
2009;33:287–97. 2013;70:56–65.
[81] Grady SA, Hussaini MY, Abdullah MM. Placement of wind turbines using [99] Shakoor R, Hassan MY, Raheem A, Rasheed N. The modelling of wind farm
genetic algorithms. Renew Energy 2005;30:259–70. layout optimization for the reduction of wake losses. Indian J Sci Technol
[82] Huang HS. Distributed genetic algorithm for optimization of wind farm 2015;8:1–9.
annual profits. In: International Conference on Intelligent Systems Applica- [100] Pookpunt S, Ongsakul W. Optimal placement of wind turbines within wind
tions to Power Systems. ISAP 2007; 2007. p. 1–6. farm using binary particle swarm optimization with time-varying accelera-
[83] Kusiak A, Song Z. Design of wind farm layout for maximum wind energy tion coefficients. Renew Energy 2013;55:266–76.
capture. Renew Energy 2010;35:685–94. [101] Turner SDO, Romero DA, Zhang PY, Amon CH, Chan TCY. A new mathematical
[84] Mosetti G, Poloni C, Diviacco B. Optimization of wind turbine positioning in programming approach to optimize wind farm layouts. Renew Energy
large windfarms by means of a genetic algorithm. J Wind Eng Ind Aerodyn 2014;63:674–80.
1994;51:105–16.