Renewable Energy 224 (2024) 120136

Contents lists available at ScienceDirect

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

Analysis of NREL-5MW wind turbine wake under varied incoming

turbulence conditions
Songyue Liu a, b, Qiusheng Li a, b, *, Bin Lu a, b, Junyi He a, b
a
Department of Architecture and Civil Engineering, City University of Hong Kong, Kowloon, Hong Kong
b
Architecture and Civil Engineering Research Centre, City University of Hong Kong Shenzhen Research Institutes, Shenzhen, 518057, China

A R T I C L E I N F O A B S T R A C T

Keywords: The offshore environment is an ideal site for wind farm development as it offers abundant wind energy resources.
Inflow turbulence generation Investigating offshore wind turbine wake under different turbulence conditions is essential for enhancing energy
Proper orthogonal decomposition generation efficiency and promoting sustainable development. This paper utilizes a synthetic turbulence method
Actuator line method
called Discretizing and Synthesizing Random Flow Generation (DSRFG) to generate four categories of turbulence
Offshore wind turbine
Wake flow analysis
environments based on International Electrotechnical Commission design requirements for offshore wind tur­
bines. A standalone NREL offshore 5-MW baseline wind turbine is then simulated using the actuator line method
coupled with large eddy simulation. Proper Orthogonal Decomposition is employed to analyze the wake char­
acteristics under different incoming turbulence conditions. For the overall wake behaviors, the flow mode, in
which some tip vortices flow back towards the hub, become more significant with higher incoming turbulence
intensity. Those vortices disturb the turbine and result in unfavorable effects. Meanwhile, when the detailed
characteristics of the wake are studied, each primary mode experiences an increase in the number of small-scale
vortices with higher incoming turbulence intensity. It results from the interaction between the wake and outer
flow, which accelerates wake recovery and benefits the downstream turbine.

1. Introduction offshore wind turbine design [5], categorizing them into four turbulence
characteristics (very high, higher, medium, and lower), with turbulence
Wind energy has been increasingly developed in recent years intensity reference values of 18%, 16%, 14%, and 12% respectively.
because of its advantages of large reserves, wide distribution, and Hence, it is necessary to investigate the wake characteristics under these
renewable characteristics. As a clean energy alternative, it does not turbulence conditions.
produce harmful emissions or contribute to the greenhouse effect, To generate different categories of incoming turbulence environ­
making it an attractive solution for reducing greenhouse gases and air ments, grids or obstacles mounted upstream of a turbine model were
pollution. According to the 2023 Global Wind Report published by the generally equipped in wind tunnel tests [6–8]. It is unfeasible to conduct
Global Wind Energy Council, the global total installed wind capacity full-scale wind turbine tests in the wind tunnels due to their limited
reached 906 GW in 2022, in which offshore wind power accounted for spaces, while the expense of such tests in the wind tunnels is relatively
only 7.1%, with a cumulative installed capacity of 64.3 GW [1]. These high. So, numerical simulation is widely utilized to study the wake
statistics highlight the significant gap between the development of characteristics of the full-scale wind turbines under turbulent conditions
offshore and onshore wind power, indicating that offshore wind power [9–11]. The Reynolds-averaged Navier-Stokes (RANS) method has been
still has promising prospects for further growth. widely used to simulate turbulent flows. It solves the flow equations by
The wake refers to the airflow formed after the wind passes through a time-averaging flow variables coupled with additional turbulence
turbine, which significantly influences the turbines located downstream models [12]. The incoming turbulence intensity can be assigned in the
[2–4]. Understanding the wake characteristics of offshore wind turbines parameter settings. For instance, Siddiqui et al. [13] employed the RANS
is crucial for enhancing power generation efficiency and achieving method coupled with the k-ε turbulence model to investigate the wake
optimal design of offshore wind farm layouts. The International Elec­ characteristics under different inflow turbulence intensities of 5%, 10%,
trotechnical Commission (IEC) has specified turbulence conditions for 15%, and 20%, respectively. However, the RANS method averages

* Corresponding author. Department of Architecture and Civil Engineering, City University of Hong Kong, Kowloon, Hong Kong.
E-mail address: [email protected] (Q. Li).

https://doi.org/10.1016/j.renene.2024.120136
Received 6 August 2023; Received in revised form 5 December 2023; Accepted 10 February 2024
Available online 14 February 2024
0960-1481/© 2024 Elsevier Ltd. All rights reserved.
S. Liu et al. Renewable Energy 224 (2024) 120136

turbulent structures over time, resulting in a relatively coarse descrip­ and simulation accuracy. Several studies have demonstrated the reli­
tion of turbulent structures [14]. Since complicated turbulent vortices ability of ALM coupled with LES for investigating wake characteristics
usually occur behind a wind turbine, a more accurate simulation of the [32–34]. The Simulator for Offshore Wind Farm Applications (SOWFA),
turbulent structures is required to reveal the flow characteristics. developed by NREL, is an open-source software for simulating the
Compared to the RANS method, Large Eddy Simulation (LES) offers aerodynamics and structural dynamics of offshore wind turbines. Within
higher accuracy in simulating turbulent structures. It distinguishes be­ this software, a solver named pisoFoamTurbine integrates the ALM and
tween large-scale eddies and small-scale turbulent effects to handle LES. It does not require precursor ABL simulation but can couple the
turbulent flows [9]. Currently, it is commonly employed to investigate DSRFG method to generate the incoming turbulence at the inlet
the wake characteristics with incoming turbulence, which enables the boundary. So, it is adopted to simulate the wake of the NREL-5MW
generation of the atmospheric boundary layer (ABL) through precursor turbine in this study.
simulations, containing information on wind velocity and turbulence In summary, a limited number of studies have focused on the simu­
intensity. The turbulence intensities generated by ABL simulations are lation of wind turbine wake under different incoming turbulence levels.
affected by various preset parameters, such as the height of the ABL Most of them encountered difficulties in generating incoming turbulence
[15], aerodynamic roughness lengths [16], and atmospheric thermal intensities with specific values, which cannot meet the IEC recommen­
stability [17]. The value of turbulence intensities can only be checked dations on turbulence conditions for designing offshore wind turbines.
after completing a precursor simulation, which is usually Therefore, this paper presents an innovative application of a synthetic
time-consuming. In this case, significant time and computational re­ turbulence method called DSRFG in simulating the wake of a typical
sources are required to achieve the desired incoming turbulence in­ wind turbine. It can generate desired incoming turbulence levels for LES,
tensity by adjusting the preset parameter multiple times. Another which ensures simulation accuracy and also satisfies the specific
approach to generate incoming turbulence in LES is to set obstacles in incoming turbulence levels recommended by the IEC. The wind turbine
the numerical simulation setup, such as a cylinder placed in front of the is modeled using pisofoamTurbine, a solver that combines the ALM and
wind turbine [18] or a cube mounted on the floor upstream the turbine LES, enabling a comprehensive evaluation of the impact of different
[11]. These structures are utilized to change the flow and generate the incoming turbulence on the wake of the turbine. To accurately study the
incoming turbulence by adjusting preset parameters, such as the number wake characteristics, the POD technique is employed to analyze the
and size of obstacles and the distance from the wind turbine. Never­ overall behaviors and detailed features of the wake, which can provide
theless, generating turbulence with precise predefined values remains guidance and suggestions for the wind turbine layout of offshore wind
challenging. farms.
Synthetic turbulence is a reliable approach for generating flows that The structures of this paper are as follows: Section 2 introduces the
satisfy specific spatial distribution and turbulence characteristics for LES theory of LES, ALM, DSRFG, and POD. Section 3 presents the basic in­
[19–23]. One method for synthesizing turbulence is to perform a fast formation about the NREL-5MW wind turbine, the settings of numerical
Fourier transform to wind speed profiles on the time series to obtain simulations, and the process of generating the four categories of inflow
power spectral and cross-spectral density. Subsequently, a triangular turbulence using the DSRFG method. In Section 4, the reliability of the
sequence based on Gaussian random numbers is constructed to synthe­ pisoFoamTurbine solver in simulating the wake characteristics of NREL-
size turbulence [24]. However, it should be noted that the turbulence 5MW turbines under the inflow turbulences is verified using two ex­
generated by this method may not satisfy the continuity equation. An amples. Section 5 employs the POD method to analyze the overall and
alternative synthetic turbulence method based on the energy spectrum detailed characteristics of the wake under different inflow conditions.
has been developed to overcome this limitation [19,20]. This method The conclusions are provided in Section 6.
can generate a random wind field that satisfies a specific wind spectrum
and turbulence intensity, which can be used as incoming turbulence for 2. Theories in numerical simulation
LES. Therefore, a spectrum-based method called the discretizing and
synthesizing random flow generation (DSRFG) technique put forward by 2.1. Large eddy simulation
Huang et al. [23] is employed in the simulation to quantity the influence
of different incoming turbulence on the wake characteristics of wind The wake of the NREL-5MW wind turbine is simulated using the
turbines in this study. Large Eddy Simulation (LES). The large-scale vortices in the flow field
To gain an in-depth understanding of wake characteristics, it is are resolved numerically, while the small-scale vortices are modeled
imperative to analyze the principal flow modes in the wake. Proper using suitable subgrid-scale (SGS) models [9]. Therefore, it can accu­
Orthogonal Decomposition (POD) is a data-driven dimensionality rately simulate large-scale turbulent structures like vortices, which
reduction technique employed to extract dominant modes from high- usually appear in the wake. In LES, the continuity equation and
dimensional data [25], which has advantages in analyzing the struc­ Navier-Stokes (N–S) equation are filtered to remove the small-scale
ture of complex flow fields [26,27]. Therefore, it is used to analyze the turbulent fluctuations, and the filtered equations are then solved using
overall behaviors and detailed features of the wake in this study. the Finite Volume Method. The Finite Volume Method discretizes the
Considering the significant proportion of support structure costs in domain into discrete control volumes and establishes conservation
the overall expenses of an offshore wind system, each wind turbine equations for properties like mass, momentum, and energy on these
should have a minimum power rating of 5 MW or higher for offshore control volumes. These equations are then solved numerically to obtain
wind systems to achieve cost-effectiveness [28]. Therefore, the NREL a numerical solution for the flow field [35]. After the filtering procedure,
offshore 5-MW baseline wind turbine (NREL-5MW wind turbine), a the continuity equation and the N–S equation [36] are expressed as
utility-scale multimegawatt wind turbine developed by the National follows:
Renewable Energy Laboratory (NREL), is selected in this research. Since ( )
∂ ∂ ( ) 1 ∂p ∂ ∂u ∂τij fi
the structure of the NREL-5MW wind turbine, especially the rotor ui + ui uj = − + ν i − + (1)
∂t ∂xj ρ ∂xi ∂xj ∂xj ∂xj ρ
component is highly complex, many small grids are usually generated
around the rotor during meshing, substantially decreasing computa­
∂ui
tional efficiency. Thus, it is necessary to simplify the wind turbine rotor. =0 (2)
∂xi
Commonly used methods for rotor simplification include the actuator
disc method (ADM) [29], the actuator line method (ALM) [30], and the where the horizontal lines on the variables in the formulas represent
actuator surface method (ASM) [31]. Among these methods, ALM is spatial filtering, ui is velocity, p is pressure, and ν is kinematic viscosity. fi
widely adopted due to its ability to balance computational efficiency

2
S. Liu et al. Renewable Energy 224 (2024) 120136

is the volume force, which can be calculated by ALM. τij is the SGS stress source term. The lift and drag force on each blade element [39]
tensor. Then, the one equation eddy (OEE) model is utilized as the SGS are expressed as follows:
model, and an extra equation [37] is expressed as follows:
1
( ) [ ] dFN = ρVrel 2 c(CL cos φ + CD sin φ)dr (10)
∂kSGS ∂ uj ∂ ∂kSGS 2
+ = (ν + νt ) − τij Sij − ε (3)
∂t ∂xj ∂xj ∂xj
1
dFT = ρVrel 2 c(CL cos φ − CD sin φ)dr (11)
where kSGS is the SGS kinetic energy, and Sij is the rate-of-strain tensor. 2
The viscous dissipation ε and sub-grid viscosity νt are expressed as where c is the chord length of the airfoil, ρ is the air density, and Vrel is
follows: the velocity of the wind relative to the blade surface. CL and CD is the
(
3/2
) coefficient of lift and drag, respectively. φ is the relative angle at which
k
ε = Cε SGS (4) the wind approaches the blade.
Δ
Due to the rotation of the actuator line within the computational
domain, the center of a blade element may not necessarily coincide with
1/2
νt = Ck kSGS Δ (5) a grid center. As a result, the velocity of the blade element must be
determined based on the velocity of the nearest grid center [30]:
where Ck and Cε are default values, which are 0.0673 and 0.93,
respectively. The SGS stress tensor τij can be calculated as follows: uij = ucell + dS⋅∇uij (12)

2
τij = − 2νt Sij + δij kSGS (6) where ucell is the grid center velocity nearest to the blade element, and dS
3 is the displacement vector from the leaf element center to the nearest
grid center.
2.2. Actuator line method
(4) Recalculation of the flow field: The flow field is updated by
The actuator line method (ALM) [30] combines blade element mo­ recalculating the velocity and pressure distributions using the
mentum theory (BEM) and computational fluid dynamics (CFD) to new volume force. If the computation achieves convergence, the
model the flow field around wind turbine blades. The ALM offers a po­ simulation proceeds to the next time step. Otherwise, steps (2)
tential advantage over traditional wind turbine modeling methods, as it and (3) are repeated until convergence.
reduces computational costs by avoiding the need for high-resolution
grids near the blade surface. This method utilizes actuator lines to 2.3. Discretizing and synthesizing random flow generation (DSRFG)
represent the blades and compute their aerodynamic loads, which can be
applied to a coarser grid, thereby improving computational efficiency to Discretizing and synthesizing random flow generation (DSRFG) is a
simulate wind turbine wake. The whole process can be divided into four synthetic turbulence method proposed by Huang et al. [23], which is a
steps. computational technique used to generate a turbulent flow field for
numerical simulations. The generated turbulence field is then used as an
(1) Initialization: The blades of a wind turbine are discretized into input for LES to study the wake characteristics with different incoming
multiple blade elements, and their aerodynamic performance is turbulence intensities. Compared to alternative methods of synthetic
evaluated using the BEM theory. turbulence, this approach can satisfy the continuity equation and
(2) Calculation of the flow field: The force of the rotor on the flow generate flow fields with specific turbulence integral scales and in­
field is added into the N–S equation as the volume force f. Then, tensities, spectra and correlations. Moreover, it overcomes the issue of
the forces at the blade elements are smooth-distributed onto the rapid energy decay in the inertial subrange and dissipation subrange. As
grid to prevent numerical oscillations using the three- a result, it can generate spatially correlated turbulent fields that conform
dimensional Gaussian distribution [30]: to the von Karman spectrum or arbitrary spectra in the three orthogonal
fε = f ⊗ ηε (7) directions.
A given continuous spectrum E(k) can be discretized into the com­
[ ( )]
1 d
2 bination of several single pieces of spectrum Em(k) [23], which is
ηε (d) = 3 3/2 exp − (8) expressed as follows:
επ ε
∑
kmax ∑
kmax kmax (
∑ )
3
where d is the distance between the initial force point on the blade E(k) = Em (k) = E(km )δ(k − km ) = νm 2 δ(k − km ) (13)
2
element and the cell center. ε is the parameter that adjusts the strength m=k0 m=1 m=1

of the Gaussian distribution function, called Gaussian width. Every

where k is the turbulence kinetic energy, and different distributions of k
element in the domain is affected by all the blade element force, so the
correspond to various types of fluctuating velocity fields. For any Em(k),
force of the flow f’ in point p [38] is defined as:
its fluctuating flow field is defined as:
∑NB ∑NE (⃒ ⃒)
f ′(p) = − fij ⋅η ⃒p − pij ⃒ (9) N [
∑ ( ) ( )]
(14)
i=1 j=1
um.i = pm,n
i cos kjm,n xj + ωm,n t + qm,n
i sin kjm,n ⋅ xj + ωm,n t
where NB is the number of wind turbine blades, NE is the number of
n=1

elements in each blade, and fij is the aerodynamic force on the jth element where pm,n and qm,n are determined by E(km); ωm,n represents the fre­
⃒ ⃒ i i
⃒ ⃒
of the ith blade. ⃒p − pij ⃒ represents the distance between p and the jth quency of the spectrum. Finally, a homogeneous and isotropic turbulent
element of the ith blade. field is synthesized as follows:
∑
kmax ∑
kmax
(3) Recalculation of the volume force: The modified volume force f’ u(x, t) = um (x, t) =
can be reevaluated based on the information of the updated flow m=k0 m=k0

field and subsequently incorporated into the N–S equations as a ∑

N
[ m,n ( ) ( )]
× p cos km,n ⋅ x + ωm,n t + qm,n sin km,n ⋅ x + ωm,n t (15)
n=1

3
S. Liu et al. Renewable Energy 224 (2024) 120136

Table 1 where, U is left singular vectors, S is singular values, and V is right

Basic properties of NREL-5MW Baseline wind turbine.
singular vectors. Based on these vectors, the following vibrations can be
Parameters Descriptions calculated [40]:
Rating 5 MW
Rotor Orientation, Configuration Upwind, 3 Blades φ(x) = V′ (17)
Drivetrain High Speed, Multiple-Stage Gearbox
Rotor, Hub diameter 126m, 3m Di = Si 2 (18)
Hub Height 90m
Cut-In, Rated, Cut-out Wind Speed 3 m/s, 11.4 m/s, 25 m/s
Cut-In, Rated Rotor Speed 6.9 rpm, 12.1 rpm

Table 2
Airfoil information of NREL-5MW wind turbine.
Airfoil type Distance from the Chord Twist Thickness (% of
hub (m) (m) (deg) the chord)

Cylinder1 2.8667 3.542 13.308 1

Cylinder1 5.6 3.854 13.308 1
Cylinder2 8.3333 4.167 13.308 1
DU40_A17 11.75 4.557 13.308 0.4
DU35_A17 15.85 4.652 11.48 0.35
DU35_A17 19.95 4.458 10.162 0.35
DU30_A17 24.05 4.249 9.011 0.3
DU25_A17 28.15 4.007 7.795 0.25
DU25_A17 32.25 3.748 6.544 0.25
DU21_A17 36.35 3.502 5.361 0.21
DU21_A17 40.45 3.256 4.188 0.21
NACA64_A17 44.55 3.01 3.125 0.18
NACA64_A17 48.65 2.764 2.319 0.18
NACA64_A17 52.75 2.518 1.526 0.18
NACA64_A17 56.1667 2.313 0.863 0.18
NACA64_A17 58.9 2.086 0.37 0.18
NACA64_A17 61.6333 1.419 0.106 0.18

2.4. Proper Orthogonal Decomposition

Fig. 2. Mesh refinement of the computational domain.

Proper Orthogonal Decomposition (POD) is a data-driven technique
that can decompose a spatiotemporal signal into products of indepen­
dent temporal and independent spatial signals [25]. Given the Table 3
complexity of wind turbine wakes, the application of POD facilitates the Classification and description of turbulence environments.
identification and description of primary dynamic features within the
Type Irefa Description
flow field so as to reduce the dimensions of the data and reveal key
information about the flow field [40]. The procedures of POD analysis A+
0.18 Designates very high turbulence environment
A 0.16 Designates higher turbulence environment
are as follows. B 0.14 Designates medium turbulence environment
A signal G(t, x) is handled based on Singular Value Decomposition C 0.12 Designates lower turbulence environment
(SVD) [40] as follows: a
Iref is the reference value of turbulence intensity and can be used to deter­
[U S V] = svd(G(t, x)) (16) mine the standard deviation parameter of turbulence (see Equation (22)).

Fig. 1. Overview of computational domain and layout of the NREL-5MW wind turbine.

4
S. Liu et al. Renewable Energy 224 (2024) 120136

Table 4 signal G(t, x) is crucial to capture the dynamic characteristics accurately.

Turbulence parameters of the Kaimal spectrum. According to the Nyquist Sampling Theorem [41], the sampling fre­
Longitudinal Lateral Vertical quency should satisfy as follows:
Standard deviation σ 0.8σ 0.5σ f ≥ 2fsignal (20)
Turbulence integral scale 8.1Λ 2.7Λ 0.66Λ
where, f is the sampling frequency and fsignal is the highest frequency in

A(t) = U • S (19)
Table 5
where, φ(x) represents the spatial characteristics, which can be called Accuracy of the simulation values compared with the target values at the wind
“mode”. Di is the energy of the ith mode. The modes are ordered based turbine position.
on their energy value from largest to smallest, i.e., the mode with the Case number (a) (b) (c) (d)
largest energy value is called “Mode 1”, and so on. A(t) represents the
Acc (Velocity) 98.5% 97.9% 98.5% 98.5%
temporal term, meaning the amplitude of the mode over time. Acc (Turbulence intensity) 91.2% 89.5% 88.8% 90.5%
Furthermore, selecting an appropriate sampling frequency for the

Fig. 3. Simulation results of longitudinal velocity and turbulence intensity profiles at inlet boundary and wind turbine position: (a) Iref = 12%; (b) Iref = 14%; (c) Iref
= 16%; (d) Iref = 18%.

5
S. Liu et al. Renewable Energy 224 (2024) 120136

Fig. 4. Comparison of the dimensionless velocity in the wake region of the NREL-5MW wind turbine between the simulation and experimental results in x, y and
z directions.

the sampled signal.

Table 6
Accuracy of the simulation results compared with experimental results.
3 Precursor simulation
Plane x/R = 0.583 x/R = 3.083 x/R = 5.083

Acc (Velocity along x direction) 95.5% 97.2% 97.3% 3. Precursor simulation

3.1. Overview of NREL-5MW wind turbine

The NREL Offshore 5-MW Baseline wind turbine (NREL-5MW wind

turbine) is a wind power generation device developed by the National
Renewable Energy Laboratory (NREL) in the United States. It offers an
efficient and dependable solution for power generation in marine

6
S. Liu et al. Renewable Energy 224 (2024) 120136

Fig. 5. Basic information of horizontal axis wind turbine (HAWT), computational domain, and wind turbine position.

rotational speed of the generator according to changes in wind speed,

enabling optimal power generation performance [44]. The basic prop­
erties and airfoil information of the NREL-5MW wind turbine [45] are
listed in Table 1 and Table 2, respectively.

3.2. Basic settings for simulation

A standalone NREL-5MW wind turbine was simulated under

different turbulence conditions. The incoming wind speed at the hub
height was set at 11.4 m/s, which is the rated speed of the turbine. The
simulation was conducted over a total duration of 1400 s. The initial
800 s allowed the wind to flow from the inlet to the outlet, ensuring the
full development of the wake. The subsequent 600 s (i.e., 120 rotation
period of the turbine) were set to analyze the wake characteristics under
long-term evolution. The simulation time step was set at 0.05 s, a choice
made to balance the computational accuracy and efficiency. The
computational domain measured 3500 × 500 × 500 m, providing suf­
ficient space for the wake to fully develop, with a distance of 200 m
between the wind turbine and the inlet boundary, as shown in Fig. 1.
The number of background cells in the x, y, and z directions was 350, 50,
and 50, respectively.
When modeling the wind turbine using the ALM method, the aero­
dynamic force on the wind turbine was incorporated into the flow field
as a volume force. As a result, there was no specific need to grid the wind
turbine itself, and the background mesh did not require excessive
Fig. 6. Mesh refinement of the HAWT computational domain.
refinement either. Therefore, mesh refinement was performed twice
around the wind turbine, as shown in Fig. 2.
environments. Specifically designed for offshore wind farms, this Since the flow through the inlet boundary was turbulent, the DSRFG
structure exhibits excellent strength and stability to ensure reliable method was employed to generate the inflow velocity field with tur­
operation even under harsh weather conditions [42]. The NREL-5MW bulence in the inlet boundary. The procedures and the characteristics of
wind turbine incorporates variable-pitch technology, allowing for wind speed and turbulence profiles are described in the next section. The
adjusting blade angles based on wind speed and direction to maximize pressure boundary condition was set as zero-gradient type, ensuring an
the utilization of available wind resources [43]. Additionally, it is accurate representation of flow behaviors near the boundaries. The
equipped with a variable-speed transmission system that adjusts the outlet velocity boundary adopted to inlet-outlet type to deal with

7
S. Liu et al. Renewable Energy 224 (2024) 120136

Fig. 7. Simulation results of longitudinal velocity, turbulence intensity profiles.

complex flow patterns, while the pressure boundary condition was involved four cases of wind fields without wind turbines using the
fixed-value type, which was set as 0. As the whole computational DSRFG method. The incoming turbulence conformed to the re­
domain was large enough, the velocity and pressure boundaries of other quirements of categories A+, A, B, and C, respectively. Vhub was set as
surfaces were applied with slip and zero-gradient type, respectively. 11.4 m/s, which is the rated wind speed of the NREL-5MW wind turbine.
zhub represented the hub height of the wind turbine, which is set at 90 m.
3.3. Generation of incoming turbulence based on DSRFG To address the potential decay of turbulence with the increase of the
distance away from the inlet, velocity sampling was performed at both
To investigate the wake characteristics of a wind turbine under the inlet boundary and wind turbine position once the wind field
varying incoming turbulence conditions, it becomes imperative to reached a stable state. This step is crucial in verifying the accuracy of
generate inflow turbulence at the inlet boundary. The techniques for turbulence generation and ensuring that the turbulence intensity profile
generating incoming turbulence are based on the DSRFG method, as at the wind turbine position still aligns with the desired curve.
introduced in Section 2.3. Following the design requirements for The simulation results for categories A+, A, B, and C are presented in
offshore wind turbines specified by International Standard No. 61400-3- Fig. 3 (a)~(d), with the turbulence intensity reference value of 12%,
1:2019 [5], the wind profile is formulated as a function of the height z, 14%, 16%, and 18%, respectively. Target values of the wind speed
which can be defined as follows: profile and turbulence intensity profile can be derived according to the
( )0.2 IEC specifications [5]. Since the energy of the wind flow is only
V(z) = Vhub
z
(21) extracted by the turbine rotor, the incoming wind speed and turbulence
zhub intensity within the rotor area are primarily focused on when studying
wake characteristics. The simulation results at the wind turbine position
where Vhub is the velocity at the hub of the wind turbine, and zhub is the from the upper to the lower side of the turbine rotor are justified by
hub height of the wind turbine. The turbulence categories are classified accuracy Acc, which is defined as follows:
as A+, A, B, and C. A detailed description of each category is listed in (
n ⃒ ⃒)
Table 3. 1∑ ⃒yi − gi ⃒
The energy spectrum of turbulence follows the Kaimal spectrum Acc = 1 − ⃒ ⃒ × 100% (24)
n i=1 ⃒ yi ⃒
[46]. Table 4 provides the standard deviation and turbulence integral
scale along the longitudinal, lateral, and vertical directions that meet the where, n is the number of the data, yi is the target value or experimental
spectral requirements. The parameters σ and Λ are defined as the tur­ value, and gi is the simulation value.
bulence standard deviation parameter and turbulence scale parameter, Table 5 indicates that the accuracy of the incoming wind speed in
respectively, which are expressed as follows: these four cases is approximately 98%, demonstrating a high level of
σ = Iref (0.75Vhub + 5.6) (22) precision. Although there is a slight reduction of the incoming turbu­
lence intensity at the wind turbine position compared to that at the inlet
{ boundary position, the accuracy remains at approximately 90% within
0.7z z ≤ 60m
Λ= (23) the rotor area, which is an acceptable range.
42m z ≥ 60m

According to the above requirements, the precursor simulation

8
S. Liu et al. Renewable Energy 224 (2024) 120136

Fig. 8. Comparison of wake velocity distributions of HAWT by simulation and experimental results.

4.1. Validation based on wake velocity distribution in uniform flow

Table 7
Accuracy of the HAWT simulation results compared with experimental results To validate the accuracy of the pisoFoamTurbine solver in simulating
along x direction. the velocity distribution within the wake region, simulation is con­
Plane x/R = x/R = x/R = x/R = x/R = x/R = ducted for the NREL-5MW wind turbine at the rated wind speed of 11.4
0.5 1.0 2.0 3.0 5.0 7.0 m/s. The wake velocity distribution is then compared with the wind
Acc (Velocity 92.7% 94.8% 96.3% 96.2% 93.3% 93.0% tunnel test results [47]. The computational domain, wind turbine posi­
along x tion, and grid resolution were established following the setup described
direction)
in Section 3.2.
Fig. 4 shows the distribution of wake velocity along the radial di­
4. Numerical validation rection of the rotor on the x/R = 0.583, 3.083, and 5.083 planes,
respectively. R is the radius of the rotor, x is the distance to the wind
As this study mainly focuses on the wake of the wind turbine, it is turbine along x-axis, and r is the distance to the hub along the radial
necessary to validate the accuracy of the wake velocity distribution direction. Each velocity component is normalized by the streamwise
simulated by the pisoFoamTurbine. It should be noted that only uniform wind speed U0, which is 11.4 m/s. The magnitude of velocity in the x-
inflow is considered in wind tunnel tests of the NREL-5MW wind turbine axis typically falls within the range of 7 m/s to 11.4 m/s, significantly
[47]. Hence, the wake velocity distribution of the NREL-5MW wind surpassing the velocities in the y and z directions that are predominantly
turbine is simulated under uniform inflow condition using piso­ below 0.5 m/s. So, the numerical validation mainly focuses on the ve­
FoamTurbine solver and compared with the results of the wind tunnel locity along x direction. The accuracy between the simulation results
test. Furthermore, an extension validation is to confirm the reliability of and experimental results is determined using Equation (24), and the
the solver under incoming turbulence conditions. Subsequently, a hor­ results are listed in Table 6. The obtained accuracy exceeds 95%, indi­
izontal axis wind turbine (HAWT) is selected for further validation since cating a high level of precision. Therefore, it is reliable for piso­
its wake distribution with incoming turbulence has been measured in FoamTurbine to accurately simulate the velocity distribution in the
wind tunnel test [8]. The HAWT model is established, and the wake wake region of the NREL-5MW wind turbine.
velocity distribution is simulated using the solver. The simulation results
are subsequently compared with the data from the wind tunnel test. The 4.2. Validation based on wake velocity distribution with incoming
details of the above two validations are described below. turbulence

Since this study aims to investigate the characteristics of wind tur­

bine wake under different initial turbulence conditions, it is necessary to

9
S. Liu et al. Renewable Energy 224 (2024) 120136

Fig. 9. Mode energy proportion under four categories of turbulence environments over a 600-s rotation period of the turbine: (a) Accumulated energy proportion of
all the modes; (b) Mode energy proportion of the first 10 modes.

verify the accuracy of pisoFoamTurbine in simulating the velocity dis­ [48]. The background grid was discretized with 62, 36, and 36 cells in
tribution of the wake with incoming turbulence. Due to lack of wind the x, y, and z directions. Subsequently, the grid was refined twice near
tunnel test results for the NREL-5MW wind turbine under initial turbu­ the wind turbine, and the grid configuration is illustrated in Fig. 6.
lence conditions, another two-blade horizontal axis wind turbine Following the settings in the cited literature [8], the longitudinal ve­
(HAWT) [8] is chosen for validation. It closely matches the NREL-5MW locity and turbulence intensity were set as the constant along the wind
turbine in its horizontal axis design. Besides, it is specifically selected for height, which is 7 m/s and 13.5%, respectively. The incoming turbu­
its turbulent inflow conditions, which resemble the conditions of this lence was generated using the DSRFG method at the inlet boundary. The
study, with a turbulence intensity of 13.5%. Reference [48] also pro­ mean wind speed profile and turbulence intensity profile are shown in
vides detailed specifications and parameters for the wind turbine, Fig. 7. The accuracy of wind speed and turbulence intensity within the
facilitating the modeling process for validation. Previous studies have rotor area is 99.7% and 95.1%, respectively. The pisoFoamTurbine
conducted wind tunnel tests on this wind turbine, measuring the wake solver was employed to simulate the behaviors of the HAWT, and the
velocity distribution under incoming speed of 7 m/s and turbulence wake velocity distribution was compared with the corresponding wind
intensity of 13%. These measurements [8] are utilized as the benchmark tunnel test results.
data for comparison. Fig. 8 illustrates the wake velocity distribution along the diameter of
The numerical simulation was conducted within a computational the turbine rotor on the x/R = 0.5, 1.0, 2.0, 3.0, 5.0, and 7.0 planes. The
domain measuring 6.2 × 3.6 × 3.6 m. The wind turbine was positioned wake velocities are presented in a dimensionless form. The accuracy of
at 0.75 m from the inlet boundary, as shown in Fig. 5. The red and blue the HAWT simulation results for each plane is listed in Table 7.
boxes in Fig. 5 indicate the parameters and airfoil sections of the wind Compared with the results of the wind tunnel test, the accuracy calcu­
turbine, respectively. Specific values can be found in the cited literature lated based on Equation (24) ranges from 92.7% to 96.3%, exhibiting a

10
S. Liu et al. Renewable Energy 224 (2024) 120136

Fig. 10. Structure and energy proportions of the first three modes over a 600-s rotation period: (a) Iref = 12%; (b) Iref = 14%; (c) Iref = 16%; (d) Iref = 18%.

high-level precision. These findings indicate the accurate capability of duration influences the properties of the captured flow field, this study
pisoFoamTurbine in simulating the wake velocity distribution of a wind employs different sampling frequencies and durations to investigate the
turbine under the incoming turbulence. overall and detailed features of the wake. Considering the rotational
speed of the NREL-5MW wind turbine is 12.1 rpm, the corresponding
5. Wake characteristics revealed by POD analysis frequency is 0.202 Hz. The minimum required sampling frequency
should be 0.404 Hz based on Equation (20). The details of sampling
In this section, the POD is utilized to analyze the overall and detailed frequency and wake analysis are described below.
characteristics of the wake. Since the choice of sampling frequency and

11
S. Liu et al. Renewable Energy 224 (2024) 120136

Fig. 11. Power spectral density analysis of the modes under four categories of turbulence environments over a 600-s period.

5.1. Analysis of the overall characteristics of the wake vorticity (counterclockwise) and negative vorticity (clockwise) are
indicated by red and blue, respectively, which are normalized from − 1
By employing a low sampling frequency and a long sampling dura­ to 1. The black arrow represents the velocity in the flow field. D and zhub
tion, it becomes feasible to capture the overall dynamic behaviors and are rotor diameter and hub height, respectively.
the large-scale structures of the wake. It can provide a preliminary The changes in the proportions of the first three modes are
analysis of the general characteristics of the wake. Subsequently, after compared, as shown in Fig. 10. When the incoming turbulence intensity
the wind field reached a stable state, the long-term evolution of the wake reference value Iref is 12%, the first three modes are denoted as Mode A,
was investigated within 800s–1400s, including 120 rotation periods of Mode B, and Mode C, respectively. It is evident that as the turbulence
the turbine. The sampling frequency was selected as 5 Hz to ensure an intensity increases, the energy proportion of mode B increases signifi­
adequate sampling frequency while minimizing data volume and cantly. For every 2% rise in Iref, its energy proportion experiences an
computational complexity. The wake velocity distribution perpendic­ approximate 5% rise, increasing from 10.9% to 28.0%. It even becomes
ular to the y-axis plane behind the wind turbine was chosen, acquiring the mode with the highest energy proportion when Iref is 16% and 18%.
3000 flow field snapshots. Conversely, the energy proportion of mode A decreases consistently,
Through the POD analysis of the original signal, valuable informa­ declining from 23.0% to 16.9%, while the proportion of mode C remains
tion can be extracted, including modes, corresponding energy pro­ relatively stable. It indicates that an increase in Iref makes mode B
portions, and corresponding amplitude over time. The modes and become more significant throughout the flow field.
energy proportion were first analyzed. The POD decomposition pro­ Mode B exhibits two distinct regions of large-scale vortices behind
duced a total of 3000 modes, which were sorted in descending order the wind turbine. One part comprises the vortices generated by the tip
based on their energy proportions. Fig. 9 (a) shows that the first 150 rotation at z/zhub about 1.8, while the other part consists of vortices
modes account for approximately 95% of the total energy. It indicates behind the hub height at z/zhub of 1. These two regions of vortices rotate
that these modes effectively contain the dominant structures and modes in opposite directions, aligning with the findings of the literature [49,
within the wake. Examining the energy proportion of the first ten modes 50]. Also, based on the arrows in Fig. 10, the velocity direction of the
shown in Fig. 9 (b), it is evident that the first three modes possess flow field in this area shows that the tip vortices do not solely flow
relatively higher energy proportions, while the remaining modes downstream. Instead, some vortices flow back towards the hub, giving
contribute less than 5% of the total energy, respectively. These three rise to large-scale vortices in this region. The rise in the energy pro­
modes represent the primary flow modes within the flow field. portion of Mode B with the increase of Iref indicates a more intense
A detailed analysis of the first three modes is conducted. It should be impact of the flow on the turbine hub. Such flow directed towards the
noted that only the magnitude and sign of the POD modes do not solely hub has adverse effects on the wind turbine. Literature [48,51] also
indicate their physical significance or influence on the flow field. When pointed out that the increase in Iref leads to amplified velocity fluctua­
various modes are compared, it is essential to consider factors such as tions, which are more inclined to generate large-scale vortices that
waveform, frequency, and other relevant information associated with impact the wind turbine blades and hub, resulting in increased fatigue
each mode. Therefore, each mode has been processed in a dimensionless loads and reduced power output of the turbine. Their conclusions are
form to comprehensively compare the displayed vortices and structures. consistent with the views in this study. Therefore, during the long-term
The vorticity and velocity of each mode are shown in Fig. 10. Positive evolution of the wake, the increase in Iref has an adverse effect on the

12
S. Liu et al. Renewable Energy 224 (2024) 120136

Fig. 12. Mode energy under four categories of turbulence environments over one rotation period of the turbine. (a) Accumulated energy of all the modes; (b) Mode
energy of the first 10 modes.

wind turbine. of total energy compared to the tailing modes, it can be concluded that
After completing the analysis of the modes and their corresponding the dominant motion modes during the long-term wake evolution are
energy proportion, a power spectral analysis was conducted to evaluate characterized by low-frequency vibrations. Fig. 10 indicates that the
the amplitude of each mode over time, providing insights into the fre­ flow patterns of these front-rank modes primarily consist of large-scale
quency characteristics and energy distribution of the wake. Fig. 11 vortices. It means that those low-frequency vibrations are related to
employs a dimensionless form on the Y axis to facilitate visualization, in the vibrations of the large-scale vortices. This finding is also supported
which the unit of power spectral density (PSD) is converted into deci­ by reference [52–54].
bels. This conversion enhances the display of the PSD and facilitates a Fig. 11 also illustrates that all the PSD curves corresponding to
better understanding of the frequency components and energy distri­ different turbulence intensities exhibit similar trends, indicating that the
bution within each temporal mode. According to the Nyquist sampling change of the incoming turbulence levels barely affects the amplitude of
theorem, the highest frequency that can be accurately analyzed is half of each mode over time.
the sampling frequency [41], which is 2.5Hz. Therefore, the PSD vari­
ation within the range of 0 Hz–2.5 Hz is displayed.
In the flow field, the PSD value of Mode 1 peaks at a frequency of 5.2. Analysis of the detailed characteristics of the wake
0.007 Hz, followed by a gradual decrease with increasing frequencies.
This indicates that the primary variation frequency of Mode 1 is low. As The rapid changes within the wake and the fine-scale structures can
the mode number increases, the peak of PSD shifts towards higher fre­ be observed in detail using a high sampling frequency and a short
quencies, indicating that the vibration frequencies of these modes are sampling time. It allows for a comprehensive understanding of the
high. Since the front-rank modes hold a significantly higher proportion complicated characteristics of the wake and the dynamic modes asso­
ciated with fast changes. One rotation period of approximately 5 s for

13
S. Liu et al. Renewable Energy 224 (2024) 120136

Fig. 13. Structure and energy proportions of the first three modes over one rotation period: (a) Iref = 12%; (b) Iref = 14%; (c) Iref = 16%; (d) Iref = 18%.

the wind turbine was selected to investigate the wake characteristics. analysis of the first ten modes, it becomes evident that only the first
The sampling frequency was set to 20 Hz. The study focused on the three modes contribute more than 5% of the total energy, respectively,
velocity distribution within the flow field, specifically in the plane while the remaining modes contribute a smaller proportion. This
perpendicular to the y-axis behind the wind turbine. A total of 2400 flow observation indicates that, within a rotation period, the dominant flow
field snapshots were obtained for analysis. patterns of the wake are relatively limited, and only the first three modes
The analysis first focuses on each mode and its energy proportion. are necessary to capture the major flow patterns during a cycle.
Contrary to the findings discussed in Section 5.1 regarding the long-term Furthermore, the change of incoming turbulence intensity hardly im­
evolution in the wake, the modes within a single rotation period exhibit pacts the energy proportion of the modes within a single rotation period.
different characteristics, as shown in Fig. 12. Specifically, the first three Fig. 13 shows the first three modes over one rotation period of the
modes account for over 95% of the total energy of the flow field. With turbine. Similar to Fig. 10, each mode has been processed in a

14
S. Liu et al. Renewable Energy 224 (2024) 120136

Fig. 14. Power spectral density analysis of the modes under four categories of turbulence environments over one rotation period.

dimensionless form to comprehensively compare the displayed vortices Mode 5 exhibit similar patterns. As the mode number surpasses 10, the
and structures. The vorticity and velocity of each mode are shown in peak of the PSD shifts towards higher frequencies, suggesting that these
Fig. 13. Positive vorticity (counterclockwise) and negative vorticity modes exhibit rapid changes in the flow. Due to their minor energy
(clockwise) are indicated by red and blue, respectively, which are proportion, these flow modes are typically of less concern. Since the
normalized from − 1 to 1. The black arrow represents the velocity of the previous analysis has highlighted the importance of the first three
flow field. D and zhub are rotor diameter and hub height, respectively. modes, it can be inferred that the predominant vibration of the wake
When the incoming turbulence intensity reference value Iref is 12%, over the short-term evolution is primarily linked to the rotation of the
the first three modes are denoted as Mode A, Mode B, and Mode C, wind turbine.
respectively. It can be observed that the energy proportion of each mode In addition, it is observed that the levels of incoming turbulence have
remains relatively stable with the increase of the Iref. Furthermore, when a negligible impact on the PSD values. All the PSD curves corresponding
investigating the structural characteristics of these modes, it becomes to different turbulence levels exhibit similar trends.
evident that at lower Iref values, the wake exhibits clear trends and
meandering patterns. However, as Iref increases, the number of small- 5.3. Differential analysis between overall and detailed characteristics
scale vortices in the wake region steadily rises, leading to an increas­
ingly unclear wake shape. Literature [55] reported that small-scale When examining the detailed characteristics within a single rotation
vortices primarily occur in regions characterized by persistent flow period, the cumulative energy proportion of the first three modes sur­
interaction. The main flow interaction in the wake exists between the passes 95%, suggesting that the first three modes are sufficient to
wake and the outer flow [56], which can accelerate wake recovery [57]. describe the flow field. Nevertheless, in the analysis of overall charac­
Hence, it can be concluded that during the short-term evolution of wake, teristics within 120 rotation periods, the first three modes only
the quantity of small-scale vortices increases with higher incoming contribute to approximately 40% of the total energy, while the cumu­
turbulence intensity, which reflects the process of wake recovery. This lative energy proportion of the first 150 modes can account for 95%. It
process is beneficial to the downstream turbine. means that 150 modes are required to well describe the flow field. The
After analyzing the mode structure and energy distribution, a power reasons for the difference need further investigation.
spectrum analysis was conducted to evaluate the amplitude of each Fig. 15 displays the flow structures of Mode 1 obtained from different
mode over time, as depicted in Fig. 14. According to the Nyquist sam­ single rotation periods when Iref is 12%. It can be seen that the flow
pling theorem, the highest frequency that can be accurately analyzed is structures of Mode 1 are different during various single rotation periods
half of the sampling frequency [41], which is 10 Hz. Therefore, the PSD (i.e., 5s). Fig. 13 has revealed that the energy proportion of Mode 1 is
variation within the range of 0 Hz–10 Hz is displayed. high in a single rotation, but Fig. 15 reflected that those flow modes do
It should be noted that Mode 1 demonstrates a distinct pattern not repeat in each single rotation period. It means that they cannot be
compared with that in the long-term evolution. A consistently high PSD simplified to one or several modes with high energy proportions when
value is observed within the 0.1 Hz–0.4 Hz frequency range, indicating a considering the long-term evolution of the wake. In contrast, their en­
predominant periodic motion pattern in the wake. Notably, the rotation ergy proportions would experience a considerable reduction over the
speed of the NREL-5MW wind turbine is 12.1 rpm, equating to long-term evolution, even though they hold substantial proportions of
approximately 0.2 Hz [45]. This suggests that the turbine rotation and energy within a single rotation period. It leads to the emergence of many
associated vibrations are the principal motion in Mode 1. Mode 2 to flow modes in the long-term evolution, each of which does not have a

15
S. Liu et al. Renewable Energy 224 (2024) 120136

the overall characteristics of the wake, the mode, where some tip
vortices flow back towards the hub, becomes more significant with
the increase of incoming turbulence intensity. The reflux region is
close to the wind turbine, which can disturb the turbine and result in
unfavorable effects. With analysis of the detailed characteristics,
there is an increase in the number of small-scale vortices in each
mode with the rise of the incoming turbulence intensity. These
vortices reflect the recovery process of the wake, which is beneficial
to the downstream turbine.
2. The change of the incoming turbulence levels barely affects the
amplitude of each mode over time, but different features are
observed during the short-term and long-term evolution of the wake.
The power spectral analysis of amplitude shows that low-frequency
vibration, which is related to large-scale vortices, plays a dominant
role in the long-term evolution of the wake. Furthermore, the vi­
bration caused by the rotation of the wind turbine plays a prominent
role in the short-term evolution of the wake.
3. The wake during long-term evolution exhibits a higher level of
complexity compared to short-term evolution. The reason is that
some detailed flow modes observed in each single rotation period are
unique and do not repeat across every period. These specific features
individually have a relatively modest impact on the wake vibration,
but their cumulative effect alters the characteristics and behavior of
the whole flow field.

In future studies, the generation process of the incoming turbulence

can be further refined to align more accurately with target values. Also,
studying a single wind turbine is insufficient for wind farm design.
Future research will encompass multiple wind turbines to investigate the
impact of varied incoming turbulence on the wakes of these turbines.

CRediT authorship contribution statement

Songyue Liu: Methodology, Writing – original draft, Writing – re­

view & editing, Software. Qiusheng Li: Conceptualization, Writing –
review & editing, Supervision, Funding acquisition. Bin Lu: Software,
Writing – review & editing. Junyi He: Methodology, Writing – review &
editing.
Fig. 15. Mode 1 at different rotation periods under the lower turbulence
environment (Iref = 12%).
Declaration of competing interest
high energy proportion but contributes to a significant cumulative en­
ergy proportion. This is the reason why the cumulative energy cannot The authors declare that they have not known competing financial
reach 95% until around 150 modes for the wake analysis of the long- interests or personal relationships that could have appeared to influence
term evolution. the work reported in this paper.
In summary, the long-term evolution of the wake is much more
complex than the short-term evolution. The reason is that the dominant
Acknowledgment
flow patterns that appear in a single rotation period do not repeat in
other periods. This greatly increases the complexity of flow patterns in
This work was carried out using the computational facilities, CityU
the long-term evolution of the wake.
Burgundy, managed and provided by the Computing Services Centre at
City University of Hong Kong (https://www.cityu.edu.hk/). The work
6. Conclusion
described in this paper was fully supported by grants from the Science,
Technology and Innovation Commission of Shenzhen Municipality
In this study, a solver that combines large eddy simulation and
(Project No: SGDX2020110309300301, JCYJ20220818101201003),
actuator line method was applied to simulate the wake velocity distri­
the National Natural Science Foundation of China (Project No:
bution of the NREL 5-MW wind turbine. The investigation focused on
51978593), and the Research Grants Council of Hong Kong (Project No:
different inflow turbulence environments based on the offshore wind
CityU 11204020).
turbine design requirements from the IEC. These environments include
very high, higher, medium, and lower turbulence levels. The DSRFG
method was employed to synthesize the incoming turbulence. The References
overall and detailed characteristics of the wake behind the turbine were
[1] M. Hutchinson, F. Zhao, Global Wind Report 2023, Global Wind Energy Council,
analyzed using the POD method. The primary findings are concluded as Bonn, Germany, 2023.
follows. [2] N. Troldborg, G.C. Larsen, H.A. Madsen, K.S. Hansen, J.N. Sørensen, R. Mikkelsen,
Numerical simulations of wake interaction between two wind turbines at various
inflow conditions, Wind Energy 14 (7) (2011) 859–876.
1. The change of the incoming turbulence levels affects the wake mode, [3] M.H.B. Ahmadi, Influence of upstream turbulence on the wake characteristics of a
leading to both favorable and unfavorable effects. With analysis of tidal stream turbine, Renew. Energy 132 (2019) 989–997.

16
S. Liu et al. Renewable Energy 224 (2024) 120136

[4] H. Meng, L. Li, J.H. Zhang, A preliminary numerical study of the wake effects on [31] W.Z. Shen, J.H. Zhang, J.N. Sørensen, The actuator surface model: a new Navier-
the fatigue load for wind farm based on elastic actuator line model, Renew. Energy Stokes based model for rotor computations, J. Sol. Energy Eng. 131 (1) (2009)
162 (2020) 788–801. 011002.
[5] I.E. Commission, Wind Energy Generation Systems-Part 3-1: Design Requirements [32] M.H. Baba-Ahmadi, P. Dong, Validation of the actuator line method for simulating
for Fixed Offshore Wind Turbines, International Electrotechnical Commission, flow through a horizontal axis tidal stream turbine by comparison with
Geneva, Switzerland, 2019. measurements, Renew. Energy 113 (2017) 420–427.
[6] C. Sicot, P. Devinant, T. Laverne, S. Loyer, J. Hureau, Experimental study of the [33] Z. Yu, X. Zheng, Q. Ma, Study on actuator line modeling of two NREL 5-MW wind
effect of turbulence on horizontal axis wind turbine aerodynamics, Wind Energy 9 turbine wakes, Appl. Sci. 8 (3) (2018) 434.
(4) (2006) 361–370. [34] C. Wu, Q. Wang, R. Yuan, K. Luo, J. Fan, Large eddy simulation of the layout effects
[7] C. Chu, P. Chiang, Turbulence effects on the wake flow and power production of a on wind farm performance coupling with wind turbine control strategies, J. Energy
horizontal-axis wind turbine, J. Wind Eng. Ind. Aerod. 124 (2014) 82–89. Resour. Technol. 144 (5) (2022) 051304.
[8] Q.A. Li, J. Murata, M. Endo, T. Maeda, Y. Kamada, Experimental and numerical [35] R. Eymard, T. Gallouët, R. Herbin, Finite volume methods, Handb. Numer. Anal. 7
investigation of the effect of turbulent inflow on a Horizontal Axis Wind Turbine (2000) 713–1018.
(part II: wake characteristics), Energy 113 (2016) 1304–1315. [36] U. Piomelli, Large-eddy simulation: achievements and challenges, Prog. Aero. Sci.
[9] J. Matsfelt, L. Davidson, Large eddy simulation: a study of clearings in forest and 35 (4) (1999) 335–362.
their effect on wind turbines, Wind Energy 24 (12) (2021) 1388–1406. [37] E. Pomraning, C.J. Rutland, Dynamic one-equation nonviscosity large-eddy
[10] A. Arabgolarcheh, S. Jannesarahmadi, E. Benini, Modeling of near wake simulation model, AIAA J. 40 (4) (2002) 689–701.
characteristics in floating offshore wind turbines using an actuator line method, [38] M. Churchfield, S. Lee, P. Moriarty, Overview of the Simulator for Wind Farm
Renew. Energy 185 (2022) 871–887. Application (SOWFA), National Renewable Energy Laboratory, 2012, p. 22.
[11] G. De Cillis, S. Cherubini, O. Semeraro, S. Leonardi, P. De Palma, The influence of [39] J.N. Sørensen, W.Z. Shen, Computation of Wind Turbine Wakes Using Combined
incoming turbulence on the dynamic modes of an NREL-5MW wind turbine wake, Navier-Stokes/actuator-Line Methodology, European Wind Energy Conference and
Renew. Energy 183 (2022) 601–616. Exhibition, France, 1999, pp. 156–159, 1999.
[12] G. Alfonsi, Reynolds-averaged Navier-Stokes equations for turbulence modeling, [40] A. Quarteroni, A. Manzoni, F. Negri, A. Quarteroni, A. Manzoni, F. Negri,
Appl. Mech. Rev. 62 (2009) 040802. Construction of RB spaces by SVD-POD, reduced Basis methods for Partial
[13] M.S. Siddiqui, A. Rasheed, M. Tabib, T. Kvamsdal, Numerical analysis of NREL differential equations, An Introduction 92 (2016) 115–140.
5MW wind turbine: a study towards a better understanding of wake characteristic [41] E. Por, M. van Kooten, V. Sarkovic, Nyquist-Shannon sampling theorem, Leiden
and torque generation mechanism, in: Journal of Physics: Conference Series, IOP University 1 (1) (2019) 1–2.
Publishing, Denpasar, Indonesia, 2016 032059. [42] B.R. Resor, Definition of a 5MW/61.5 M Wind Turbine Blade Reference Model,
[14] Y.T. Wu, F. Porte-Agel, Simulation of turbulent flow inside and above wind farms: Sandia National Lab.(SNL-NM), Albuquerque, NM (United States), 2013.
model validation and layout effects, Boundary-Layer Meteorol. 146 (2) (2013) [43] Y. Nam, C. Im, Control system design of NREL 5MW wind turbine, Journal of the
181–205. Korean Solar Energy Society 32 (5) (2012) 31–40.
[15] A. Al Sam, R. Szasz, J. Revstedt, An investigation of wind farm power production [44] M. Moness, A.M. Moustafa, Hybrid modelling and predictive control of utility-scale
for various atmospheric boundary layer heights, J. Energy Resour. Technol. 139 (5) variable-speed variable-pitch wind turbines, Trans. Inst. Meas. Control 42 (9)
(2017) 051216. (2020) 1724–1739.
[16] Y.T. Wu, F. Porté-Agel, Atmospheric turbulence effects on wind-turbine wakes: an [45] J. Jonkman, S. Butterfield, W. Musial, G. Scott, Definition of a 5-MW Reference
LES study, Energies 5 (12) (2012) 5340–5362. Wind Turbine for Offshore System Development, National Renewable Energy Lab,
[17] L.L. Tian, Y.L. Song, N. Zhao, T.G. Wang, W. Zhong, Iop, Numerical investigations America, 2009, pp. 1–75.
of wind turbine wakes under neutral and convective atmospheric stability [46] J.C. Kaimal, J. Wyngaard, Y. Izumi, O. Coté, Spectral characteristics of surface-
conditions, in: 7th Conference on Science of Making Torque from Wind (TORQUE), layer turbulence, Q. J. R. Meteorol. Soc. 98 (417) (1972) 563–589.
2018 072044. Milan, Italy. [47] S.H. Kang, K.W. Ryu, An experimental study on wake flow-field of NREL 5 MW
[18] M.H.B. Ahmadi, Z. Yang, On wind turbine power fluctuations induced by large- wind turbine model, Journal of the Korean Society for Aeronautical & Space
scale motions, Appl. Energy 293 (2021) 116945. Sciences 45 (2) (2017) 85–91.
[19] K. Kondo, S. Murakami, A. Mochida, Generation of velocity fluctuations for inflow [48] Q.A. Li, J. Murata, M. Endo, T. Maeda, Y. Kamada, Experimental and numerical
boundary condition of LES, J. Wind Eng. Ind. Aerod. 67 (1997) 51–64. investigation of the effect of turbulent inflow on a Horizontal Axis Wind Turbine
[20] P. Hémon, F. Santi, Simulation of a spatially correlated turbulent velocity field (Part I: power performance), Energy 113 (2016) 713–722.
using biorthogonal decomposition, J. Wind Eng. Ind. Aerod. 95 (1) (2007) 21–29. [49] F. Viola, G.V. Iungo, S. Camarri, F. Porté-Agel, F. Gallaire, Prediction of the hub
[21] P. Batten, U. Goldberg, S. Chakravarthy, Interfacing statistical turbulence closures vortex instability in a wind turbine wake: stability analysis with eddy-viscosity
with large-eddy simulation, AIAA J. 42 (3) (2004) 485–492. models calibrated on wind tunnel data, J. Fluid Mech. 750 (2014) R1.
[22] F. Mathey, D. Cokljat, J.P. Bertoglio, E. Sergent, Assessment of the vortex method [50] R. Ashton, F. Viola, S. Camarri, F. Gallaire, G.V. Iungo, Hub vortex instability
for large eddy simulation inlet conditions, Progress in Computational Fluid within wind turbine wakes: effects of wind turbulence, loading conditions, and
Dynamics, Int. J. 6 (1–3) (2006) 58–67. blade aerodynamics, Physical Review Fluids 1 (7) (2016) 073603.
[23] S.H. Huang, Q.S. Li, J.R. Wu, A general inflow turbulence generator for large eddy [51] W. Tian, A. Ozbay, H. Hu, A wind tunnel study of wind loads on a model wind
simulation, J. Wind Eng. Ind. Aerod. 98 (10–11) (2010) 600–617. turbine in atmospheric boundary layer winds, J. Fluid Struct. 85 (2019) 17–26.
[24] J. Nathan, C. Masson, L. Dufresne, Near-wake analysis of actuator line method [52] Z.B. Li, X.H. Liu, X.L. Yang, Review of turbine parameterization models for Large-
immersed in turbulent flow using large-eddy simulations, Wind Energy Science 3 Eddy Simulation of wind turbine wakes, Energies 15 (18) (2022).
(2) (2018) 905–917. [53] P.A. Brugger, C.D. Markfort, F. Porté-Agel, Field measurements of wake
[25] A. Chatterjee, An introduction to the proper orthogonal decomposition, Curr. Sci. meandering at a utility-scale wind turbine with nacelle-mounted Doppler lidars,
78 (7) (2000) 808–817. Wind Energy Science Discussions 2021 (2021) 1–22.
[26] L.H.O. Hellström, T.C. Hohman, A.J. Smits, POD analysis of the structure of [54] G. Dong, J. Qin, Z. Li, X. Yang, Characteristics of wind turbine wakes for different
vertical axis wind turbine wakes, J. Wind Eng. Ind. Aerod. 237 (2023) 105403. blade designs, J. Fluid Mech. 965 (2023) A15.
[27] D. Bastine, B. Witha, M. Wächter, J. Peinke, Towards a simplified dynamic wake [55] G. Ruetsch, M. Maxey, The evolution of small-scale structures in homogeneous
model using POD analysis, Energies 8 (2) (2015) 895–920. isotropic turbulence, Phys. Fluid. Fluid Dynam. 4 (12) (1992) 2747–2760.
[28] W. Musial, S. Butterfield, A. Boone, Feasibility of floating platform systems for [56] F. Porté-Agel, M. Bastankhah, S. Shamsoddin, Wind-turbine and wind-farm flows: a
wind turbines, in: 42nd AIAA Aerospace Sciences Meeting and Exhibit, 2004, review, Boundary-Layer Meteorol. 174 (1) (2020) 1–59.
p. 1007. [57] T. Maeda, Y. Kamada, J. Murata, S. Yonekura, T. Ito, A. Okawa, T. Kogaki, Wind
[29] I. Fejtek, L. Roberts, Navier-Stokes computation of wing/rotor interaction for a tilt tunnel study on wind and turbulence intensity profiles in wind turbine wake,
rotor in hover, AIAA J. 30 (11) (1992) 2595–2603. J. Therm. Sci. 20 (2) (2011) 127–132.
[30] J.N.r. Sørensen, W.Z. Shen, Numerical modeling of wind turbine wakes, J. Fluid
Eng. 124 (2) (2002) 393–399.

17