Professional Documents
Culture Documents
by
SIMENG LI
August, 2014
i
CASE WESTERN RESERVE UNIVERSITY
SCHOOL OF GRADUATE STUDIES
Simeng Li
Committee Chair
J. IWAN D. ALEXANDER
Committee Member
JAIKRISHNAN R. KADAMBI
Committee Member
PAUL BARNHART
Committee Member
DAVID H. MATTHIESEN
Date of Defense
July 2, 2014
ii
Contents
Chapter 1 Introduction ......................................................................................................................... 1
1.1 Background ............................................................................................................................. 1
1.2 Wake measurements and wake model .................................................................................. 5
1.3 Wind farm layout optimization .............................................................................................. 6
1.4 Dissertation Outline................................................................................................................ 7
Chapter 2 Literature Review ................................................................................................................. 8
2.1 Wake models .......................................................................................................................... 8
2.1.1 Analytical Wake models ................................................................................................. 8
2.1.2 Numerical Wake model ................................................................................................ 10
2.1.3 Multiple wake models .................................................................................................. 12
2.2 Wind farm performance evaluation models ........................................................................ 13
2.3 Wind farm layout optimization problem .............................................................................. 15
Chapter 3 Objectives .......................................................................................................................... 18
3.1 Wind array performance evaluation .................................................................................... 18
3.2 Wind farm layout optimization ............................................................................................ 19
Chapter 4 Wake models ..................................................................................................................... 21
4.1 Kinematic Wake Models ....................................................................................................... 23
4.1.1 Jensen wake model ...................................................................................................... 24
4.1.2 Larsen model ................................................................................................................ 25
4.1.3 Frandsen analytical model............................................................................................ 26
4.2 Field Wake Models ............................................................................................................... 27
4.2.1 Ainslie wake model ....................................................................................................... 28
4.2.2 Three-dimensional field model .................................................................................... 29
4.3 Wake added turbulence models .......................................................................................... 31
Chapter 5 Methodology...................................................................................................................... 34
5.1 Large wind array performance evaluation model (LWAP) ................................................... 34
5.1.1 Multiple wake model .................................................................................................... 36
5.1.2 Effect of the wind array on the atmospheric boundary layer ...................................... 42
5.2 Wind array layout optimization model (WALOM)................................................................ 46
5.2.1 Wind array configuration set up................................................................................... 46
5.2.2 Wind data and distribution........................................................................................... 49
iii
5.2.3 Cost function ................................................................................................................ 55
5.2.4 Turbine power .............................................................................................................. 56
5.2.5 Wake effect evaluation................................................................................................. 57
5.2.6 Genetic Algorithm (GA) ................................................................................................ 58
Chapter 6 Results: wind farm layout evaluation model ..................................................................... 63
6.1 Wind speed evaluations at Horns Rev when wind direction is along turbine rows ............. 63
6.2 Turbine power evaluations at Horns Rev: wind direction parallel to turbine rows ............. 69
6.3 Power output predictions for turbines in the row at the Horns Rev and Nysted for a
representative wind speed and variable wind directions ................................................................ 73
Chapter 7 Results: wind farm layout optimization ............................................................................. 79
7.1 Extension of Mossetti’s approach ........................................................................................ 79
7.1.1 Case 1: Constant unidirectional wind ........................................................................... 80
7.1.2 Case 2: Constant wind speed with an equal probability variable wind direction ........ 85
7.1.3 Case 3: Variable Wind Speed with Variable Wind Direction ........................................ 87
7.1.4 Different Spacing Limits for Unidirectional Wind ......................................................... 89
7.1.5 Different Area Sizes for Unidirectional Wind ............................................................... 92
7.1.6 Constant Wind Speed with Variable Wind Direction and a circular site area .............. 95
7.1.7 Comparison of optimized layouts using Jensen wake model and Ainslie wake model 97
7.2 Horns Rev wind farm layout optimization.......................................................................... 100
7.2.1 Weibull data for 12 wind direction sectors ................................................................ 103
7.2.2 Wind data for 16 wind direction sectors .................................................................... 105
7.2.3 Wind data for 72 wind direction sectors .................................................................... 107
Chapter 8 Conclusions and future work ........................................................................................... 110
Bibliography ........................................................................................................................................ 113
iv
Tables
TABLE 5.1 WIND DATA EXAMPLE 51
TABLE 5.2 WEIBULL FACTORS FOR DIFFERENT WIND DIRECTIONS [77]. 54
TABLE 6.1 MAPE OF THE COMPUTED NORMALIZED WIND VELOCITY USING OBSERVED DATA FOR TWO WIND
DIRECTIONS AND TWO WIND SPEEDS REPORTED IN [60] AT HORNS REV. 69
TABLE 6.2 MAPE OF THE COMPUTED NORMALIZED TURBINE POWER USING COMPUTED PREDICTIONS AND
ACTUAL OBSERVATIONS AT HORNS REV [58]: WIND DIRECTION PARALLEL TO TURBINE ROWS. 72
TABLE 6.3 RMSD OF THE COMPUTED NORMALIZED POWER FOR VARIOUS WIND DIRECTIONS AND A WIND
SPEED OF 8 M/S AT HORNS REV WIND FARM [60] 78
TABLE 6.4 RMSD OF COMPUTED NORMALIZED POWER FOR VARIOUS WIND DIRECTIONS AND A WIND SPEED
OF 8 M/S AT NYSTED WIND FARM [60] 78
TABLE 7.1 RESULTS FROM PREVIOUS STUDY AND CURRENT STUDY: REPORTED AND RECOMPUTED 83
TABLE 7.2 RESULTS FROM PREVIOUS STUDY AND CURRENT STUDY: REPORTED AND RECOMPUTED 86
TABLE 7.3 RESULTS FROM PREVIOUS STUDY AND CURRENT STUDY: REPORTED AND RECOMPUTED 89
TABLE 7.4 TURBINES DISTRIBUTIONS IN OPTIMIZED LAYOUTS FOR ISOTROPIC WIND AND ROUND AREA 96
TABLE 7.4 VESTAS V80 THRUST COEFFICIENT AND POWER AS A FUNCTION OF WINS SPEED 101
TABLE 7.5 WIND DISTRIBUTION FOR HORNS REV 102
TABLE 7.6 TURBULENCE INTENSITIES FOR VARIABLE WIND SPEEDS 102
TABLE 7.7 LAYOUT PERFORMCANCE OF OPTIMIZATION RESULTS 109
v
Figures
FIGURE 1.1 AERIAL VIEW FROM THE SOUTHEAST OF WAKE CLOUDS AT HORNS REV ON FEBRUARY 12, 2008
[11]. 2
FIGURE 4.1 WAKE PROFILE DOWNSTREAM A TURBINE 21
FIGURE 4.2 TURBINES TAKEN INTO CONSIDERATION WHEN CALCULATING ADDED TURBULENCE 32
FIGURE 5.1 OVERLAPPED WAKES WHERE 𝐬𝟏=7D 38
FIGURE 5.2 WAKE DECAY CONSTANT, 𝒌′, AS A FUNCTION OF UPSTREAM TURBINE WIND SPEED DEFICIT 39
FIGURE 5.3 COMBINATION COEFFICIENT, C, AS A FUNCTION OF NORMALIZED DOWNSTREAM DISTANCE 40
FIGURE 5.4 NORMALIZED WIND VELOCITY CALCULATED USING MULTIPLE WAKE MODEL AND THE JENSEN
WAKE SINGLE MODEL. 41
FIGURE 5.5 PREDICTED FREE STREAM WIND SPEEDS AT TURBINE HEIGHT FOR THE HORNS REV WIND FARM
LAYOUT 45
FIGURE 5.6 WIND FARM BOUNDARY SET UP 47
FIGURE 5.7 HORNS REV ARRAY TURBINE LAYOUT COORDINATES 48
FIGURE 5.8 HORNS REV TURBINES RANKING FOR NORTH WIND 48
FIGURE 5.9 HORNS REV TURBINES RANKING FOR WEST WIND 49
FIGURE 5.10 MEAN WIND SPEED FREQUENCY DISTRIBUTION (WEIBULL SHAPE FACTOR, K, AND SCALE 52
FIGURE 5.11 WIND DIRECTION ROSE 53
FIGURE 5.12 COST OF WIND FARM VS. NUMBER OF TURBINES 55
FIGURE 5.13 VESTAS V80 2MW TURBINE POWER OUTPUT AND THRUST COEFFICIENT VS. WIND SPEED [78] 57
FIGURE 5.14 GENETIC ALGORITHM PROCESS 61
FIGURE 5.15 FLOWCHART OF GENETIC ALGORITHM OPTIMIZATION PROCESS 62
FIGURE 6.1 HORNS REV LAYOUT: CASE 1 OF 270° AND 7D SPACING, CASE 2 OF 222° AND 9.4 D SPACING 64
FIGURE 6.2 HORNS REV EVALUATION WIND DIRECTION 270° AND WIND SPEED 8.5 M/S +/- 0.5 M/S 65
FIGURE 6.3 HORNS REV EVALUATION WIND DIRECTION 270° AND WIND SPEED 12 M/S +/- 0.5 M/S 66
FIGURE 6.4 HORNS REV EVALUATION WIND DIRECTION 222° AND WIND SPEED 8.5 M/S +/- 0.5 M/S 67
FIGURE 6.5 HORNS REV EVALUATION WIND DIRECTION 222° AND WIND SPEED 12 M/S +/- 0.5 M/S 68
FIGURE 6.6 POWER CURVE FOR THE TURBINE AT HORNS REV 70
FIGURE 6.7 TURBINES POWER AT CASE 1 FOR WIND SPEED AT 8M/S AND DIRECTION 270° AT HORNS REV 70
FIGURE 6.8 TURBINES POWER AT CASE 1 FOR WIND SPEED AT 10M/S AND DIRECTION 270° AT HORNS REV 71
FIGURE 6.9 TURBINES POWER AT CASE 2 FOR WIND SPEED AT 8M/S AND DIRECTION 222° AT HORNS REV 71
FIGURE 6.10 HORNS REV ARRAY. EXACT ROW (ER=270°) OF TURBINES [60] 74
FIGURE 6.11 NYSTED ARRAY. EXACT ROW (ER=278°) OF TURBINES [60] 74
FIGURE 6.12 NORMALIZED POWER AT HORNS REV FOR THE FREE STREAM WIND SPEED OF 8 ± 0.5 M/S:
COMPARISON OF MODELS WITH OBSERVATIONS 75
FIGURE 6.13 NORMALIZED POWER AT NYSTED FOR THE FREE STREAM WIND SPEED OF 8 ± 0.5 M/S:
COMPARISON OF MODELS WITH OBSERVATIONS 76
FIGURE 7.1 WIND FARM AREA 82
FIGURE 7.2 WIND DISTRIBUTION FOR CASE 3 82
FIGURE 7.3 FITNESS VALUE OF DIFFERENT NUMBER OF TURBINES FOR CASE 1. 83
FIGURE 7.4 TURBINES PLACEMENT OF FOUR STUDIES FOR CASE 1: (A) MOSSETTI ET AL. [25] (B) GRADY ET AL.
[27] (C) MARMIDIS ET AL. [28] (D) MITTAL ET AL. [29] (E) WALOM 84
FIGURE 7.5 FITNESS VALUE OF DIFFERENT NUMBER OF TURBINES FOR CASE 2 85
FIGURE 7.6 TURBINES PLACEMENT FOR CASE 2: (A) MOSSETTI ET AL. [25] (B) GRADY ET AL. [27] (C) MITTAL ET
AL. [29] (D) WALOM 86
vi
FIGURE 7.7 FITNESS VALUE OF DIFFERENT NUMBER OF TURBINES FOR CASE 3 87
FIGURE 7.8 TURBINES PLACEMENT FOR CASE 3: (A) MOSSETTI ET AL. [25] (B) GRADY ET AL. [27] (C) MITTAL ET
AL. [29] (D) WALOM 88
FIGURE 7.9 FITNESS VALUES OF DIFFERENT SPACING LIMITS 90
FIGURE 7.10 EFFICIENCIES OF DIFFERENT SPACING LIMITS 90
FIGURE 7.11 OPTIMAL PLACEMENTS OF FOUR SPACING LIMITS FOR 40 TURBINES 91
FIGURE 7.12 FITNESS VALUES OF DIFFERENT AREA SIZES 93
FIGURE 7.13 EFFICIENCIES OF DIFFERENT AREA SIZES 93
FIGURE 7.14 OPTIMAL PLACEMENT OF FOUR AREA SIZES FOR 40 TURBINES 94
FIGURE 7.15 FITNESS VALUE AS A FUNCTION OF TURBINE NUMBER 95
FIGURE 7.17 OPTIMAL TURBINES LAYOUTS FOR A CIRCULAR SITE AREA 96
FIGURE 7.18 OPTIMAL TURBINES LAYOUTS FOR 48 TURBINES WITH CASE 1 UNIFORM ONE DIRECTION (FROM
NORTH TO SOUTH) WIND USING JENSEN WAKE MODEL AND AINSLIE WAKE MODEL. 98
FIGURE 7.19 OPTIMAL TURBINES LAYOUTS FOR 48 TURBINES WITH CASE 2 FOR VARIABLE WIND DIRECTION
AND A CONSTANT WIND SPEED USING JENSEN WAKE MODEL AND AINSLIE WAKE MODEL. 99
FIGURE 7.20 OPTIMAL TURBINES LAYOUTS FOR 48 TURBINES WITH CASE 3 FOR VARIABLE WIND DIRECTION
USING THE JENSEN WAKE AINSLIE WAKE MODELS. 99
FIGURE 7.21 WIND DIRECTION DISTRIBUTION FOR 12 DIRECTION SECTORS 103
FIGURE 7.22 OPTIMIZED LAYOUT FOR CASE 1 104
FIGURE 7.23 ANNUAL WIND POWER ROSE WITH 12 DIRECTION SECTORS 104
FIGURE 7.24 WIND DIRECTION DISCRETIZATION FOR 16 DIRECTION SECTORS 106
FIGURE 7.25 ANNUAL POWER COMPARISON OF 16 DIRECTION SECTORS 106
FIGURE 7.26 WIND DIRECTION DISCRETIZATION FOR 72 DIRECTION SECTORS 107
FIGURE 7.27 OPTIMIZED LAYOUT FOR 72 DIRECTION SECTORS 108
vii
Nomenclature
𝐴 Rotor disc area
𝐴𝑛 The nth wake area
𝐴𝑟𝑛 Rotor area of the nth turbine
𝐴𝑐 The Weibull scale factor
𝐴𝑗 The Weibull scale factor of the jth wind direction
𝑎 Turbine induction factor
𝑏 Wake lateral width
𝐶 Combination factor
𝐶𝑇 Wind turbine thrust coefficient
𝐶𝑡 The distributed thrust coefficient
𝑐1 Constants in Larsen wake model
𝑐𝑚𝑤 The relative mean wind speed in the wake
𝑐𝑤𝑓 The flow speed deficit in the infinitely large wind farm
cost Total cost of the wind farm
𝐷 Turbine rotor diameter
𝐷𝑒𝑓𝑓 The effective rotor diameter
𝐷𝑀 Centerline velocity deficit
𝐷𝑚𝑖 The initial centerline velocity deficit
𝐷𝑟 The expanded downstream rotor diameter
𝐷𝑠 Empirical distance constant
𝐷𝑤 Wake width
ℎ Height of the internal boundary layer
ℎ𝐻 Turbine hub height
𝐼𝑎 The ambient turbulence intensity
𝐼𝑎′ The ambient turbulence intensity in a turbine wake
𝐼𝑇 Maximum center wake turbulence intensity
𝐼𝑤 Turbulence intensity in the wake
𝐾 The Weibull shape factor
𝐾𝑗 The Weibull shape factor of the jth wind direction
𝑘 Wake decay constant
𝑘′ Wake decay coefficient
𝑁 Number of turbines in the wind farm
𝑃 Power output of a turbine
𝑃𝑎𝑣𝑎𝑖𝑙 Power available from the wind
𝑃𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑 Observed turbine power
𝑃𝑝𝑟𝑒𝑑𝑖𝑒𝑐𝑡𝑒𝑑 Predicted turbine power
𝑃𝑡𝑜𝑡𝑎𝑙 Total wind farm power
viii
𝑅9.5 The wake radius at 9.5 rotor diameters downstream the turbine
𝑅𝑤 Rotor wake radius
𝑠 Normalized downstream distance by turbine rotor diameter
𝑠𝑖 Normalized downstream distance from the ith turbine
𝑠𝑐 Crosswind turbine spacing
𝑠𝑑 Downwind turbine spacing
𝑠𝑓 Turbine spacing
𝑈′ The relative wake velocity deficit
𝑈1 Free stream velocity at turbine height in the internal boundary layer
𝑈𝑐 Normalized centerline velocity deficit
𝑈𝑖𝑛𝑖𝑡𝑖𝑎𝑙 The initial wake velocity deficit
𝑈0 Free stream wind velocity
𝑈0𝑗 Free stream wind velocity of the jth wind direction
�
𝑈 The mean wind speed
∆𝑈 Velocity deficit in the wake
𝑢 Wind velocity in a turbine wake
𝑢𝑖 Wind velocity calculated by the wake model in the ith turbine wake
𝑢𝑖′ Wind velocity calculated by the single wake model in the ith turbine wake
𝑢𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑 Observed wind speed
𝑢𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 Predicted wind speed
𝑢∗0 Friction velocity in the atmospheric boundary layer
𝑢∗1 Friction velocity in the internal boundary layer
𝑥0 The position of the rotor respected to the applied coordinate system
𝑧0 Offshore roughness
𝑧00 Wind farm roughness
𝛼 Constant related to the trust coefficient
𝛽 Wake expansion parameter
ε𝑣 Eddy viscosity
𝜃𝑗 The jth wind direction
𝜅 Von Karman Constant
𝜌 Air density
𝜎𝑈 The standard deviation of the wind speed in the wake
𝜎𝑦 Standard deviation of wind velocity in y direction
𝜎𝑧 Standard deviation of wind velocity in z direction
ix
Abbreviations
CFD Computational Fluid Dynamics
ENDOW EfficieNt Development of Offshore WindFarms
ECN Energy Research Centre of the Netherlands
ER Exact Row
EWTS European Wind Turbine Standards
FLaP Farm Layout Program
GA Genetic Algorithms
GH Garrad Hassan
IBL Internal Boundary Layer
KAMM Karlsruhe Atmospheric Mesoscale Model
LWAP Large Wind Array Performance Evaluation Model
MAPE Mean Absolute Percentage Error
MC Monte Carlo
NTUA National Technical University of Athens
RGU Robert Gordon University
RMSD Root Mean Square Deviation
SAR Synthetic Aperture Radar
SODAR Sonic Detection and Ranging
WALOM Wind Array Layout Optimization Model
WAsP Wind Atlas Analysis and Application Program
WFOG Wind Farm Optimization using a Genetic Algorithm
x
Acknowledgements
Many people have contributed in one way or another to the completion of this work
and, while I cannot name them all, I wish to express my deep gratitude to each of them.
My first gratitude must go to my advisor, Dr. J. Iwan D. Alexander. He patiently
provided the vision, encouragement and advice necessary for me to proceed through the
doctoral program and complete my dissertation. I want to thank Iwan for his unflagging
encouragement and serving as a role model to me as a junior member of academia. He has
been a strong and supportive adviser to me throughout my graduate school career. In
addition, he has always given me a great freedom to work independently.
Special thanks to my committee, Professors Jaikrishnan R. Kadambi, Paul Barnhart
and David H. Matthiesen for their support, guidance, helpful suggestions and patience
throughout the course of my research. Their guidance has served me well and I owe them
my great appreciation.
I am heartily thankful to Hui Yi, for her help, support and encouragement
throughout my pursuit for the doctoral degree. I feel fortunate to have met you and cherish
the time we spent together in US.
Also, thanks to Professor Joseph Prahl for his constant help and encouragement on
my study and research. Thank you, Professor Bo Li, for your advice on my dissertation and
journal papers. I would also like to thank my colleagues, Yng-Ru Chen, Ying Chen, Xinyou Ke
and department assistant Sheila Campbell. I enjoy the time studying and working with you.
Finally, I would also like to express my gratitude to my parents for their care and
support. Also thanks to my friends Jinxia Guo, Hua Zhou, Lin Chen and Zhe Yang for their
support and encouragement.
This work was partially supported by the Department of Energy (Award Numbers
EE0000275 and DE-EE0005379). I would also like to acknowledge support from Case
Western Reserve University’s Great Lake Energy Institute.
xi
WIND ARRAY PERFORMANCE EVALUATION MODEL FOR LARGE WIND
FARMS AND WIND FARM LAYOUT OPTIMIZATION
Abstract
by
SIMENG LI
The grouping of wind turbines in arrays introduces two major issues: (1) reduced power
production caused by wake wind speed deficits and (2) increased dynamic loads on the
blades caused by higher turbulence levels. Depending on the layout and local wind
conditions, the drop in power production of downstream turbines can easily reach 40% of
the upstream turbines in fully developed wake conditions. These power drops across arrays
arise due to wake wind speed deficits. Even when averaged over different wind directions,
drops in power production of 8% (onshore arrays), and 12% (offshore arrays) have been
recorded. In this dissertation, a large wind array performance evaluation model (LWAP) to
evaluate wake effects in large wind farms is developed. The model accounts for multiple
wake interactions and the effect on the vertical wind profile in the atmosphere boundary
layer by the wind farm itself. The model predicts wind speed deficits at each turbine and for
specific turbine power curves and assesses power for individual turbines and for the entire
wind farm. The calculation method converges within seconds for a large wind farm
evaluation. To assess the efficacy of the wake model, measured wind speed deficits and
turbine power deficits along two wind directions and wind turbine rows in the Horns Rev
xii
wind farm were compared with deficits calculated using the model. The mean absolute
average in wind turbine power evaluation. Case studies predicting row-wise power deficits
of turbines arrays in Horns Rev and Nysted wind farms on multiple wind directions were
compared to observations. LWAP exhibits the same accuracy on power deficit evaluation as
with the CFD based models such as WindFarmer, WakeFarm and NTUA and performs better
than the WAsP Park model. The computing time to process an entire full wind farm (e.g.,
Horns Rev) is on the order of a few seconds, significantly less than the CFD based models. In
evaluate and optimize wind array performance for real wind farm site. Results of optimized
wind array layouts are obtained and analyzed on case studies of multiple wind distributions
conditions and site conditions. It is found that the optimized results are affected by factors
such as wind distribution, wind data resolution, wake model and wind farm site conditions.
xiii
Chapter 1 Introduction
1.1 Background
Wind energy is the fastest growing source of electricity and one of the fastest
growing markets in the world. At the end of 2011, worldwide wind power capacity reached
238 GW, doubled in three years [1, 2]. Wind power is renewable, clean, worldwide and
using little land [3]. In the year of 2010, it was reported that 430 TWh was generated by
wind power, which is about 2.5% of worldwide electricity usage [4, 5]. It is expected to
reach 3.35% by 2013 and 8% by 2018 [6, 7]. For the U.S. to reach 54 GW of offshore wind
energy by 2030 [8], 1000’s of wind turbines (typical offshore array sizes today range from 40
-100 turbines) will need to be installed along the Atlantic, Pacific, and Great Lakes coasts.
The cost of these developments must be competitive with traditional on-shore generation,
Wind turbines, which transform wind power into electricity, are usually grouped into
large wind farms, also called wind arrays, for reason of economies, such as lower
transportation, installation, maintenance and land costs. However, grouping turbines causes
a reduction of the power produced due to wake effects. When wind turbine extracts energy
from the wind, it produces a wake of wind velocity deficit downwind the turbine, so that the
power extracted by downstream turbines is reduced. Nowadays, a large wind farm may
consist of several hundred wind turbines, so that wake effects are unavoidable. Recent
studies found that the average power losses due to wake in a wind array are in the order of
1
10-20% [8]. In addition, the effect of increased fatigue loads of wind turbines operating in
Figure 1.1 Aerial view from the Southeast of wake clouds at Horns Rev on February 12, 2008 [11].
Figure 1 shows a famous photograph of the Horns Rev wind farm in Denmark. It
illustrates the wake by wind turbines and wake interactions in the large wind farm. During a
2
previous study [12] for this wind case, power losses of wind turbines downstream in the
The ability to accurately quantify power losses associated with wind turbine wakes is
an important aspect of wind farm design analysis. One source of uncertainty in estimating
these losses is the effects of interacting wakes within the wind farm. Flow in the wake from
levels that can adversely affect the performance of wind turbines situated downstream.
Accurate prediction of wind velocity deficit and wind turbine power deficit downstream of
wind turbines is crucial to the evaluation of wind farm layout. Specifically this reduction in
performance is manifested by subpar power output due to decreased wind speeds as well
aspects of turbine placement within a wind farm and how that placement affects the wakes
and, thus, the expected power losses from a given wind distribution. Reducing wake losses,
or even reduce uncertainties in predicting power losses from wakes, contributes to the
overall goal of reducing the cost of wind-farm operation while maximizing power produced.
Accurate prediction of wind velocity deficit downstream of the wind turbines is crucial to
The spatial configuration of wind turbines in a large scale wind power plant will
affect the overall performance of the plant and, thus, be an important contributing factor to
cost-effectiveness over the long-term. The grouping of turbines in arrays introduces two
major issues: (1) reduced power production caused by wake velocity deficits and (2)
3
increased dynamic loads on the blades caused by higher turbulence levels. Depending on
the layout and local wind conditions, the drop in power production of downstream turbines
can easily reach 40% of the upstream turbines in fully developed wake conditions. These
power drops across arrays arise due to wake velocity deficits. Even when averaged over
different wind directions, drops in power production of 8% (onshore arrays), and 12%
The land requirement for wind turbines is roughly 10 hectares (2.78 acres) per MW.
To date, with a few exceptions, most of the rationale behind spacing is intuitive, e.g., Patel
[14] proposed 8-12 rotor diameters windward 1.5-3 diameters crosswind. This was
suggested to be inefficient by Ammara [15] who proposed a denser staggered siting scheme
that would produce the same power but use less land. The placement of wind turbines in
large wind turbine arrays or wind fields will, thus, affect the overall power generation
characteristics of the wind field. If turbines in large arrays are not sited to attempt to
maximize performance while keeping within other practical site-particular constraints that
may be imposed (for example, with respect to wake interference between turbines and
with respect to the prevailing winds) then sub-par performance of the individual turbines
within the field may result. A good example of this is the Horns Rev (160 MW) wind field off
the coast of Denmark. Certainly, one of the most heavily studied wind field Horns Rev has
exposed limitations of current methods that are employed to determine turbine placement
4
1.2 Wake measurements and wake model
straight forward under certain controlled laboratory conditions, i.e., scaled wind turbine
models in wind tunnels or small turbines in large wind tunnels is more complicated in the
field as wakes are spatio-temporally variable phenomena shifting with the direction of the
wind and not amenable to meteorological measurements on a long term basis without
downstream of the prevailing wind direction. The Danish wind turbine group at RISO has
made the most advances in wake measurements. Three meteorological towers erected in
the vicinity of eleven turbines at the Vindeby offshore farm have permitted simultaneous
measurements of wind speed in the free stream and wake for several wind directions [20].
were also obtained by Barthelmie et al. [21] using ship-borne Sonic Detection and Ranging
(SODAR). Types of wake measurements were also discussed by Barthelmie et al. [8]. Wake
measurements have also been obtained from satellite SAR measurements by Christiansen
Wind turbine wakes have been studied for two decades and various models have
been developed. These models can be divided into two main categories, namely, analytical
wake models and computational or numerical wake models. An analytical wake model
computational wake models, fluid flow equations, whether simplified or not, must be solved
5
to obtain the wake velocity field. Analytical models have advantages from the viewpoint of
evaluating and designing wind farm performance because of its simplicity and
As wakes develop downstream, they interact with the atmospheric boundary layer
as well as with other wakes and are also affected by variable surface terrain. The extent of
the wake and the way in which the wake evolves is dependent on a number of factors.
These include the wind conditions (speed and turbulence intensity) entering the wind field
(the so-called inlet conditions), the terrain, the interaction of the wake with the
atmospheric boundary layer [23], the turbulence developed in the wake (wake-added
turbulence) and the characteristics of the turbine (size, blade design, hub height, etc.).
At present, in existing wind arrays, turbines are often organized in identical rows
that are separated by a convenient spacing, normally of 6-10 rotor diameters [24]. As the
spacing between turbines increases, the wake losses decrease, but results in higher inter-
turbine cabling and land costs. The problem then is to balance these competing effects by
determining the distance from shore and the turbine spacing that results in lowest possible
wake effects.
Several attempts have been made to optimize turbine placement in wind arrays and
showed that irregular arrays result in a higher energy layout than regular grids [25-37]. One
of the earliest was by Mossetti et al. [25] who developed optimal placing schemes based on
6
constrain placement to maximize power output, minimize surface area occupied by the
wind farm and minimize cost or maximize asset lifetimes as well as factoring in spatially
variable operating points for individual or groups of turbines. Mossetti [25] also applied
Jensen wake model [26] to evaluate the turbine wake. Many relevant researchers that used
the similar method to optimize this problem in the wind array can be found in the literature
[27-31]. However, the models in these studies for optimal wind turbine placement
proposed to date are strikingly simple and lack a robust wake model capable of accounting
power curves according to turbulence intensity etc., are not accounted for.
The dissertation is organized as follows: A review of previous work on wind turbine wake
modelling and wind farm layout optimization is presented in Chapter 2. A brief review of
optimization methods is also given. Chapter 3 defines and discusses the objectives of this
dissertation. Wind turbine wake and wind farm related models form the core part of the
work and the optimization modeling and are discussed in Chapter 4. Chapter 5 describes the
methodology of the wind array performance modeling. In Chapter 6, the wind array
performance evaluation model is verified against actual data obtained from Horns Rev wind
array and Nysted wind array. The application of the optimization model to two wind farm
layout case studies is described in Chapter 7. In addition, factors affecting the optimization
are analyzed. Finally, the conclusions and ideas for future work are presented in Chapter 8.
7
Chapter 2 Literature Review
This chapter provides the literature review and relevant background for this
dissertation. Section 2.1 reviews wake models that have been widely used in wind turbine
and wind farm researches. In section 2.2, models for wind farm performance evaluation and
wind farm design analysis are reviewed. Section 2.3 provides the background for the wind
Wake models can be divided into two main categories, namely, analytical wake
computational wake models, fluid flow equations, whether simplified or not, must be solved
Analytical models are first introduced by Lanchester and Betz [38, 39], who derived
the principles of conservation of mass and momentum of the wind flow through an
Jensen wake model [26] treated the wake behind the wind turbine as a turbulent
wake which ignores the contribution of vortex shedding that is significant only in the near
wake region. The wake model is thus derived by conserving momentum downstream of
8
wind turbine. The velocity in the wake is given as a function of downstream distance and it
is assumed that the wake expands linearly downstream of the wind turbine. Jensen also
proposed that when two wakes interact, the resultant kinetic energy deficit is equal to the
sum of the kinetic energy deficits of the individual wakes at that point.
The work of Katic et al. [40] expanded the previous work of Jensen and assumes the
wake velocity profile is constant or a ‘top hat’ profile. This assumption is justified by the fact
that the purpose of the model is to estimate the energy content within the wind field as
seen by downwind turbines rather than accurately describing the spatial variation velocity
field. Notably, the Katic model is currently the basis of the wake model used in WAsP [41].
WAsP is a wind climate and turbine power prediction software developed by Risø and is
widely used.
Frandsen’s kinematic model based on work completed by Larsen et al. [42], and
currently included in the European Wind Turbine Standard II, is based on the ‘classical’ wake
theory outlined by Schlichting [43]. This model proposes a semi-analytical solution for
predicting the velocity deficit within a wake in which a set of simple empirical relations are
used to predict the turbulence intensity and the turbulence length scale. It assumes an
axisymmetric wake within which the velocity deficit decays with downstream distance to
the power of 2/3 and the turbulence intensity decays to the power of 1/3. It also suggests
that the wake width increases with the downstream distance to the power of 1/3.
Ishihara et al. [44] developed an analytical wake model by taking the effect of
turbulence on the rate of recovery into account. They used similarity theory for the velocity
profile and defined wake recovery as a function of ambient turbulence and turbine
9
generated turbulence. They calculated results for both offshore and onshore conditions and
also at both high loading and low loading of wind turbine. These results compared well with
experimental data obtained using a 1/100 scale model of Mitsubishi MWT-1000 wind
turbine in a wind tunnel. The scale model used surface roughness models upstream of the
wind turbine to simulate onshore conditions and a smooth upstream surface to simulate
offshore conditions.
Werle [45] proposed a three part wake model: an exact model for the inviscid near
wake region; Prandtl’s turbulent shear layer mixing solution for the intermediate wake and
a far wake model based on the classical Prandtl/Swain axisymmetric wake analysis. No
comparisons of the model with actual data have been published to date.
Lissaman’s kinematic wake model [46] predicts the effects of individual turbine
wakes, based on self-similar velocity deficit profiles. The work involved both experimental
and theoretical components related to co-flowing jets. The results give a full definition of
the velocity profile within a wake and facilitate the prediction of the velocity deficit for a
given wake radius. The growth of the wake depends on the ambient, free stream turbulence,
the shear generated turbulence and the turbulence created by the turbine. The maximum
velocity deficit at each downwind position is found via a control volume momentum
balance, with the initial velocity deficit calculated using the thrust coefficient of the turbine.
There are a number of numerical wake models that range in complexity, and are
used by the wind turbine industry. Early work by Templin [47] and Newman [48] described
10
the effects of turbines as distributed roughness elements and were developed further by
Bossanyi et al. [49], Frandsen [50] and Emeis and Frandsen [51]. These models are useful
when predicting the effects of large wind farms on wind flow [52]. In another type of wake
model wind turbines are modeled as roughness elements [50, 52]. Frandsen combined the
drag from turbine with surface drag to get the total drag. The limitation of these types of
models is that the calculated total roughness is independent of wind direction and these
models are best suited for predicting overall effects of large wind farms on wind
characteristics.
Crespo et al. [52] carried out an extensive survey of different modeling methods for
wind turbine wakes. Apart from surveying various analytical wake models (discussed above)
she reported the computational wake model UPMWAKE to be one of the best after
calculate the flow at every position throughout the wake field. The original work of this type
was completed by Sforza et al. [53] and describes a wake using the linearized conservation
of momentum equation in the direction of the free stream flow. The wake is assumed to
have constant advective velocity, constant eddy diffusivity and a wake shape described by a
parabolic approximation.
Other field models have been developed by Taylor [54], Liu et al. [55], Crespo et al.
[56], Ainslie [57] and, more recently, Magnusson [58]. These models involve solution of a
simplified version of the Reynolds averaged Navier–Stokes flow equations. They employed
11
an ‘eddy viscosity’ turbulence model to model the turbulent mixing contributions from the
One multiple wake calculation method was described by Mossetti et al. [25].
Multiple wakes were accounted for by simply assuming that the kinetic energy deficit of a
mixed wake was equal to the sum of energy deficits. The velocity downstream of n turbines
where 𝑢 is the wind velocity for the turbine in the wake, 𝑈0 is the free stream wind velocity,
This approach to accounting for multiple wakes was also applied in other studies
[27-31]. However, Li et al. [32] showed that predictions with this method were not in
Another method was developed by Frandsen et al. [15] who presented a wake
expansion model developed by taking the control volume analysis of the momentum flow.
This model was used in WAsP engineering model for the wake evaluation [41]. The wind
𝐴𝑛 𝐶𝑇 𝐴𝑛+1
𝑈𝑛+1 = 1 − �1 − 𝑈𝑛 �1 − � 𝐴𝑟𝑛 � 2.2
𝐴𝑛+1 2 𝐴𝑛
12
where 𝐶𝑇 is the thrust coefficient, 𝐴𝑛 is the nth wake area and 𝐴𝑟𝑛 is the rotor area of the
nth turbine.
The limitation of this model is that it applies only to a single row of equally spaced
turbines aligned parallel to the wind direction. To describe real wind farms an extension of
the approach to account for an irregular wind array layout is needed. In addition, any
evaluation of multiple wake effects on the performance of the wind farm needs to consider
a representative description of the distribution of expected wind speeds and directions for a
given geographic location, i.e., a large number of discrete wind directions and wind speeds.
Finally, the required computational time must to be sufficiently low to allow for efficient
In this section, four models for the wind farm performance evaluation model are
introduced. They are the Wind Atlas Analysis and Application Program (WAsP), the GH
WindFarmer, the ECN’s WAKEFARM and the NTUA model [41, 59-65].
WAsP is based on a linearized model used in the European Wind Atlas. The WAsP
program [41] is a series of models that developed from site measured wind data at a
generalized wind climate and hence is restricted to location specific descriptors of the flow
climate. In terms of wind farm modelling, the Park wake model [40] is used in the
commercial version. A new wake model with ‘Mosaic tile’ is being developed for use within
WAsP which is described by Rathmann et al. [59]. The program utilizes the observed wind
data by fitting it to a two parameter Weibulll distribution. The main advantage of the WAsP
13
program is that it is fast and robust. It is reported that for running a full wind farm
simulation as Horns Rev wind farm, WAsP takes the order of minutes [60]. However, in the
Garrad Hassan (GH) WindFarmer is a CFD based model [61]. It applies an empirical
representation of the wind turbine wake developed by Ainslie [57]. Empirical expressions
are applied to model turbulence intensities in the turbine wake and the superposition of
individual wakes. In addition, for very large wind farms, the change of the vertical wind
profile as a result of the presence of wind turbines is modeled [62]. The computation time
for a full wind farm simulation for this model was reported as minutes [60].
ECN’s WAKEFARM model is developed from the UPMWAKE code and was developed
by the Universidad Polytecnica de Madrid [63]. This model is similar to the GH WindFarmer
model, but is extended to 3D. The wake model is computed using a 3D parabolized Navier-
Stokes code and applies a k-epsilon turbulence model. The parabolisation of equations
leads to an efficient calculation procedure, but for the near wake, an empirical velocity
profile is required to for initialization. The axial pressure gradients are assumed as external
forces and enforce the flow to decelerate. Because it computes a 3D flow filed for the
simulation of wind farms such as Horns Rev, several hours are needed to converge [60].
The NTUA CFD model is developed by National Technical University of Athens and is
based on the 3D Reynolds averaged incompressible Navier-Stokes equations [64, 65]. The
model applies the k-epsilon turbulence closure model and accommodates wind turbines
embedded in the grid as momentum sinks representing the thrust force applied on the
14
rotor disk. This model needs much more computation than models discussed above. It is
reported that it requires a time scale of days for the Horns Rev wind farm simulation [60].
The spatial configuration of wind turbines in a large scale wind power plant will
affect the overall performance of the plant and, thus, be an important contributing factor to
its cost-effectiveness over the long-term. The placement of wind turbines in large wind
turbine arrays or wind fields will affect the overall power generation characteristics of the
wind field. If turbines in large arrays are not sited to attempt to maximize performance
while keeping within other practical site-particular constraints that may be imposed (for
example, with respect to wake interference between turbines and with respect to the
prevailing winds) then sub-par performance of the individual turbines within the field, and,
The land requirement for wind turbines is roughly 10 hectares (2.78 acres) per MW.
To date, with a few exceptions, most of the rationale behind spacing is intuitive e.g., Patel
[17] proposed 8-12 rotor diameters windward 1.5-3 diameters crosswind. This was
suggested to be inefficient by Ammara [18] who proposed a denser staggered siting scheme
that would produce the same power but use less land.
Most approaches to placement of wind turbine within large arrays are based on
simplified wake models that account for different levels of interaction between turbines. In
recent work, particularly that by Frandsen, there have also been attempts to bring in more
meteorological based models [15] (in this case the Karlsruhe Atmospheric Mesoscale Model,
15
KAMM) to assist in assessing the efficacy of given wind turbine array configurations. When
compared to Horns Rev data, this was found to lead to improvements in predictive ability
but the reader is cautioned that that this approach may not translate well to different
wind farms. The first (published) attempt was made by Mossetti [25] who used genetic
algorithm for optimization. Mossetti’s idea was to develop optimal placing schemes based
constrain placement to maximize power output, minimize surface area occupied by the
wind farm and minimize cost or maximize asset lifetimes as well as factoring in spatially
variable operating points for individual or groups of turbines. The key ‘physical’ ingredient
in Mossetti’s model was the wake model. He used utilized Jensen’s model [26]. Three
The same problem was tackled by Grady et al. [27] who improved upon Mossetti’s
analysis by accounting for non-uniform and variable winds. From their study it is apparent
that there is great sensitivity to the assumptions made in the analysis. Grady showed that
Mossetti’s results were not optimum and gave improved results. However, there are some
Marmidis [28] attempted the problem of optimal placement using a Monte Carlo
(MC) method for optimization. He analyzed only one case out of three cases analyzed by
Mossetti and Grady. The biggest problem with his adaptation of the MC approach is the size
16
Wan et al. [34] improved previous works by using a Weibull function to describe the
strategy algorithm. They achieved the optimization problem which maximized the expected
Mittal et al. [29] developed a code Wind Farm Optimization using a Genetic
Algorithm (WFOG) for optimizing the placement of wind turbines in large wind farms by
using Jensen wake model [26] and the Fuga wake model developed by Ishihara et al. [44].
Mittal also used a refined grid for the possible turbine position. It allowed more flexibility in
the placement of wind turbines. It was reported that the Fuga wake model estimated the
Chen et al. [37] focused on optimizing the wind farm layout of turbines with
different hub height and under the uncertainty with landowner’s financial and noise
concerns. Their work improved transparency-of-information that can potentially affect the
negotiation process between developers and landowners during early wind farm planning
17
Chapter 3 Objectives
The overall objective of this research is to create a simulation tool suitable for
characterization of the in-situ operating environment within an offshore wind farm. The
tool will supply quantitative information necessary for wind turbine and wind farm design,
and power production estimates, and also optimize wind array layout to maximize power
production.
Accurately predicting the wind and is a necessary part of assessing the power
production potential of a given layout of turbines in the wind array area and is crucial for
the success of any wind farm project. For multi-turbine arrays, this includes modeling of
wind distribution, power curve and wake losses caused by one or more upwind turbines.
Reducing wake losses, or even reduce uncertainties in predicting power losses from wakes,
contributes to the overall goal of reducing the cost of wind-array operation while
model which is fast and suitable for the large wind farm design. A multiple wake model was
developed by only considering the wake of the nearest upstream turbine and its operating
conditions which are affected by other upstream turbine wakes. The model was developed
from the flow momentum theory of the Jensen wake model [26], see also Chapter 4. This
approach requires much less calculation than two multiple wake models discussed above
18
and can be easily applied to regular or irregular wind array layout for multiple wind
directions.
The models for optimal wind turbine placement proposed to date are strikingly simple
and lack a robust wake model capable of accounting for the effects of wakes from neighboring
turbines. In addition, changes in operating points on the turbine power curve according to
turbulence intensity etc. are also not accounted for. The wind distribution forwarded by
Mossetti et al. [25] only considered three different wind speeds and assumed an
oversimplified angular distribution. The potential for improvement is high if only by applying
actual observed wind data with different real site and applying simultaneous consideration
of how factors such as wind distribution, wake model, optimization algorithms and site area
The wind farm layout optimization analyzes turbine positioning, power production,
turbine layout and installation and operating costs in a wind array. The goal is to maximize
the power per unit cost. In this study, a comprehensive model will be developed in which
turbines’ positioning is optimized, evaluated and analyzed under simulated various real
wind farm circumstance. Factors affecting the optimization and evaluation such as wind
distribution resolution, wake model, wind farm site limitation and turbine type will be
The objective is to provide an offshore wind array optimization tool suitable for
characteristic of any offshore array site. It analyzes the performance of wind turbines within
19
the array for given external wind distributions and optimizes the array layout to maximize
array power production per unit cost. It is expected to find some patterns of turbine
placement that can obtain more power on various wind farm sites.
20
Chapter 4 Wake models
well as with other wakes and are also affected by variable surface terrain. The extent of the
wake and the way in which the wake evolves is dependent on a number of factors. These
include the wind conditions (speed and turbulence intensity) entering the wind field (the so-
called inlet conditions), the terrain, the interaction of the wake with the atmospheric
boundary layer [15, 66], the turbulence developed in the wake (wake-added turbulence)
and the characteristics of the turbine (size, blade design, hub height, etc.).
As wind passes through the region swept by the blades, over the nacelle and past
the tower, the energy extracted by the wind turbine reduces the wind speed. Figure 4.1
shows an idealization of the turbine wake. The near wake behavior is different from the far
wake region.
21
The wind speed measured immediately downstream of a turbine is significantly
lower than the free stream velocity. At near hub-height, the air in the wake is more
𝑈0 − 𝑈(𝑥)
𝑈′ = 4.1
𝑈0
where 𝑈(𝑥) is the velocity in the wake at a distance x downstream from the originating
turbine and 𝑈0 is the free-stream velocity. The wake velocity 𝑈(𝑥) and 𝑈0 are obtained
from meteorological data measured at fixed points in and around the wind field. Another
𝜎𝑈
𝐼𝑤 = 4.2
𝑈�
� is the mean wind speed (in the wake) and U is the standard deviation of the wind
where 𝑈
speed in the wake. Turbulence intensity will affect wind turbine performance and different
𝐼𝑤 can result in different wind turbine power curve characteristics for the same wind speed
[66]. Compressibility as wind moves through the plane of the rotor is negligible and the
velocity reduction across the rotor plane results in a downstream conically expand wake.
The wake continues to expand downstream and, for flat terrain, interacts with the ground
when the wake radius equals the turbine hub height. Frandsen et al. [15] reported that this
momentum from the free stream occurs due to the initially large velocity gradient and as
22
Wind turbine wakes have been studied for two decades and various models have
been developed. These models can be divided into two main categories, namely, kinematic
wake models (analytical wake models) and field wake models (computational or numerical
wake models). A kinematic wake model characterizes the velocity in a wake by a set of
analytical expressions whereas in field wake models, fluid flow equations, whether
In what follows past work on wake models relevant to wind turbines is discussed.
The models range from simple linear wake models that simply account for downstream
attenuation of wind speed in the near wake region to reattainment of the free stream
speed in the far wake, to more sophisticated computational fluid dynamics based models
that can account for wake turbulence and in some cases uneven topography.
Kinetic wake models are developed from the momentum equation to model the
velocity deficit in the wake behind a turbine. The wake descriptions usually do not consider
the near wake region (less than two turbine diameters distance downwind a turbine). They
also do not account for the change in turbulence intensity in the wake. Thus they have to be
combined with a turbulence model if it is required to account for values of the turbulence
intensity throughout the wind farm. Kinetic wake models are simple and computationally
economic in that they can be easily implemented within large scale for calculations of large
23
4.1.1 Jensen wake model
The so-called Jensen model is one of the oldest and simplest wake models
and developed by N.O. Jensen [26]. It has been used in several studies that employ a
wake model in algorithms that attempt to optimize the cost per unit power by seeking the
optimizing placement of wind turbines within a given area. He treated the wake behind the
wind turbine as a turbulent wake which ignores the contribution of vortex shedding that is
significant only in the near wake region. The wake model is thus derived by conserving
momentum downstream of wind turbine The velocity in the wake is given as a function of
downstream distance and it is assumed that the wake expands linearly downstream of the
𝐷𝑤 = 𝐷𝑟 (1 + 2𝑘𝑠) 4.3
1 − �1 − 𝐶𝑇
𝑢 = 𝑈0 [1 − ] 4.4
(1 + 2𝑘𝑠)2
where 𝐶𝑇 is turbine thrust coefficient, k is Wake Decay Constant and s=x/D is relative
downstream distance. The value of k is generally taken to be 0.075 for land cases, and 0.05
1−𝑎
𝐷𝑟 = 𝐷� 4.5
1 − 2𝑎
Katic et al. [40] expanded the previous work by Jensen, describes the wake velocity
profile as constant or ‘top hat’ profile which is justified by the fact that the purpose of the
24
model is to estimate the energy content within the wind field as seen by downwind turbines
rather than accurately describing the spatial variation velocity field. Notably, the Katic
model is currently the basis of the wake model used in WAsP. WAsP is a wind climate and
turbine power prediction software developed by Risø. It is also included in Garrad Hassan
The model developed by G.C. Larsen [42], also known as the EWTS-II model (used in
velocity profile is assumed and Prandtl’s mixing length theory [43] is used to get a closed
form solution. The flow is further assumed to be incompressible, stationary and of no wind
boundary layer equations [67], of which the last one is capable of resolving the double dip
in the velocity deficit profile of the near wake. The following equations for the, rotor wake
radius 𝑅𝑤 and the axial velocity deficit in the wake (∆𝑈)1 are obtained
1
35 5 1 1
𝑅𝑤 (𝑥) = � � (3𝑐12 )5 (𝐶𝑇 𝐴(𝑥 + 𝑥0 ))3 4.6
2𝜋
3
𝑈∞ 1 3 1 35 10 1
−
(∆𝑈)1 (𝑥, 𝑟) = − (𝐶𝑇 𝐴(𝑥 + 𝑥0 )−2 )3 [𝑟 2 �3𝑐12 𝐶𝑇 𝐴(𝑥 + 𝑥0 )� 2 − � � (3𝑐12 )5 ]2 4.7
9 2𝜋
25
Two unknown constants, 𝑐1 is respectively related to the Prandtl mixing length and
𝑥0 is the position of the rotor which respect to the applied coordinate system. Following
5 1
𝐷𝑒𝑓𝑓 2 105 −2 5
𝑐1 = � � � � (𝐶𝑇 𝐴 𝑥0 )−6 4.8
2 2𝜋
9.5𝐷
𝑥0 = 4.9
2𝑅
( 𝐷 9.5 )3 − 1
𝑒𝑓𝑓
1 + �1 − 𝐶𝑇
𝐷𝑒𝑓𝑓 = 𝐷� 4.10
2�1 − 𝐶𝑇
And 𝑅9.5 is the wake radius at 9.5 rotor diameters downstream the turbine is taken
from [30]
estimate wake behavior across an entire wind farms rather than individual turbines. It is
based on conservation of the momentum deficit in the wake. The model distinguishes three
different wake regimes: a single wake regime, two neighboring interacted wake flow
regimes and a wake flow regime that is in balance with the atmospheric boundary layer.
26
In single wake case, the velocity deficit is determined as
1 𝐴0
𝑢= 𝑈0 ± �0.25 − 0.5𝑈02 𝐶𝑇 4.13
2 𝐴
where A0 is the turbine rotor swept area immediately downstream of the rotor, and A is the
area of the wake. The expansion of the area of the wake for the single wake case is given by
𝐷 1
= 3/2 4.14
𝐷0 (𝛽 + 𝛼𝑠𝑓 )1/3
where
1 1 + �1 − 𝐶𝑇
𝛽= ∙ 4.15
2 �1 − 𝐶𝑇
and α is a constant related to the thrust coefficient, which describes the initial wake
In the case of multiple wakes, the wakes are divided in several sections each having
a constant but different velocity deficit. In order to calculate the mean wind speed over the
rotor area, a semi-linear method is applied. For details see the paper by Rathmann et al.
[59].
calculate the flow at every position throughout the wake field, by solving the Reynolds-
averaged Navier-Stokes equations with a turbulence model for closure. The following are
27
4.2.1 Ainslie wake model
The Ainslie wake model is a two-dimensional field model [57]. The model assumes
that the wake profile is axisymmetric and assumes a Gaussian profile at the two rotor
diameters downstream from the turbine. The flow is considered to be incompressible with
no external forces. Beyond the first two diameters, the gradients of mean quantities in the
axial direction are neglected as they will be much less than the gradients in the radial
direction [57]. The flow can then be described with a two-dimensional Reynolds equation in
𝜕𝑈 𝜕𝑉 𝜖𝑣 𝜕𝑈 𝜕 2𝑈
𝑈 + 𝑉 = ( + 𝑟 2) 4.16
𝜕𝑥 𝜕𝑟 𝑟 𝜕𝑟 𝜕𝑟
𝜕𝑈 1 𝜕𝑉
= − �𝑟 + 𝑉� 4.17
𝜕𝑥 𝑟 𝜕𝑟
here 𝜖𝑣 is the eddy-viscosity and used for closure. These two equations are solved
numerically starting from the near wake with an empirical wake profile. The initial
𝐼𝑎
𝐷𝑚𝑖 = 𝐶𝑇 − 0.05 − (16𝐶𝑇 − 0.5) 4.18
1000
where 𝐶𝑇 is the wind turbine’s thrust coefficient (a function of the upstream wind speed)
The wake width 𝑏 which increases with downstream distance is related to the thrust
3.56𝐶𝑇
𝑏=� 4.19
8𝐷𝑀 (1 − 0.5𝐷𝑀 )
28
Given a wake width 𝑏 and a Gaussian shape by the wind velocity in the wake U is of
the form
𝑈 2
1− = 𝐷𝑀 𝑒 −3.56(𝑟/𝑏) 4.20
𝑈0
here 𝑈0 is the free stream velocity and 𝑟 is the distance from the centerline.
The effects of turbulence and turbine operation (described in 𝐶𝑇 ) are included in the
model. A lower turbulence intensity leads to a slower wake recovery. Higher 𝐶𝑇 leads to a
stronger near wake. Software packages that apply the Ainslie wake model include WindPRO,
GH WindFarmer, FLaP.
The Ainslie model was adapted by Anderson [68] who developed a simplified
solution. He substituted equation (4.17) and equation (4.20) into equation (4.16) and
obtained
𝑈𝑐 = (1 − 𝐷𝑀 ) 4.22
The system (4.16) - (4.17) now reduces to a first order differential equation (4.21)
which can be solved efficiently by a simple numerical integration scheme. In our study, we
applied this simplifying method and decreased computing time two orders of magnitude.
29
The ECN model is based on the UPMWAKE model developed by Crespo et al. [69].
The model is a 3D parabolized Navier-Stokes code for the far wake using a k-𝜖 turbulence
model. The near wake model is solved by momentum theory and some empirical
corrections. The increase of momentum at infinity is subtracted from the undisturbed wind
where a is designated to the axial induction factor in the rotor plane. Similar to Jensen wake
model, the velocity decrease is assumed to be constant with the “expanded” rotor diameter
𝐷𝑟
1−𝑎
𝐷𝑟 = 𝐷� 4.24
1 − 2𝑎
𝐷𝑟
𝜎𝑧,𝑦 = 4.26
2𝐷
during the ENDOW project [63] based on a previous axisymmetric model by Magnusson et
al. [64]. Initial data required to start the 3D-NS calculations are the velocity and turbulence
30
intensity profiles in the atmospheric boundary layer upstream of the rotor. The
computational domain includes the rotor of the wind turbine, which is approximated by
means of a semipermeable disk to simulate the pressure drop across a real rotor disk. The
computational time is much longer for this model compare to other model [70].
intensity when used for wake calculations. In our work, we used the Frandsen
The model assumes the following: if the distance separation between two wind
turbines in a wind farm is more than 10 rotor diameters, wake effects on turbulence
intensity can be neglected. If the separation is less than 10 rotor diameter, the wake effects
on the added turbulence have to be calculated following equations. Wake added turbulence
1
𝑁 𝑚
𝐼𝑒𝑓𝑓 = �(1 − 𝑝𝑤 𝑁)𝐼𝑎𝑚 + 𝑝𝑤 𝑁 � 𝐼𝑇𝑚 (𝑠𝑖 )� 4.27
𝑖=1
where m is the Wohler curve exponent depending on the material of the structural
31
𝐼𝑎 depends on the considered load case
1
𝐼𝑇 = � + 𝐼𝑎2 4.28
1.5 + 0.3𝑠𝑖 �𝑈0
The wake of a turbine that is upstream another turbine from the point of view of the
the possible number of nearest and next nearest neighboring turbines is 8, see also Figure
4.2.
Figure 4.2 Turbines taken into consideration when calculating added turbulence
Inside large wind farms, the wind turbines tend to generate their own ambient
turbulence. Thus, when (a) the number of wind turbines from the considered unit to the
32
“edge” of the wind farm is more than 5, or (b) the spacing in the rows perpendicular to the
predominant wind direction is less than 3 rotor diameters from each other, a different value
1
Ia′ = 2 + I2 + I �
��Iw a a 4.29
2
0.36
𝐼𝑤 = 4.30
𝑠𝑟 𝑠𝑓
1 + 0.2�
𝐶𝑇
The values sr and sf are the relative spacings in the rows and between the rows of
turbine in wind array
33
Chapter 5 Methodology
This chapter describes the Wind farm performance evaluation and layout
optimization model used in subsequent chapters. The motivation for developing the model
wind array to supply quantitative information necessary to determine optimal turbine array
layout and even guide wind plant control/operation and power production estimates. In
previous chapters, the development of model approach that can provide quantitative
characterization of the in-situ operating environment throughout the array with sufficient
resolution to capture the essential dynamics within the array to quantify performance have
been presented and discussed. In the chapter, the wind array configurations are simulated,
wind data or wind distribution is processed to describe the annual average wind condition
and wake models are used to evaluate wake losses. In addition, a new multiple wake model
and a wind farm roughness model were developed by measured data from Horns Rev wind
farm and Nysted wind farm. It only needs a simple computation scheme for any complex
multiple wakes case, but its result is promising. An optimization model is also created in a
Matlab code and applies Matlab “Global Optimization Toolbox” to optimized wind array
In this section, a new model for large wind array performance evaluation to evaluate
wake effects in large wind farms is developed. To distinguish it from other models it is
34
referred to as the LWAP or the LWAP model. The LWAP applies a single wake model such as
the Jensen wake model or the Ainslie wake mode, and accounts for multiple wake
interactions and the effect on the vertical wind profile in the atmosphere boundary layer by
the wind farm itself. The LWAP predicts wind speed deficits at each turbine and for specific
turbine power curves and assesses power for individual turbines and for the entire wind
farm. The calculation method converges within seconds for a large wind farm evaluation.
• Apply wind data (speed, direction and turbulence intensity) over the wind
farm site.
• Place the wind turbines in the wind farm and calculate the new free stream
wind velocity due to the increased surface roughness in the atmospheric boundary
• Input this new free stream velocity into a combination of a single wake
model and a multiple wake model to evaluate the wake deficits in the wind farm.
The single wake models are introduced in Chapter 4. Here the multiple wake model
and the model for the evaluation of the effect by the wind array on the atmospheric
boundary layer are developed. A Matlab code was developed for this simplified model that
is suitable for wind farm evaluation. In order to save computational time for the Ainslie
implicit model, a simplified solution for the model by Anderson [68] was used, and wind
speed deficits in the field of the single wake were calculated and saved as a data base for
using for wake deficits evaluation. The code requires much less computational time than
35
CFD based wind farm evaluation models, but the accuracy is not sacrificed. For comparative
purposes both the Jensen wake model and the Ainslie wake model were used as the single
wake model to predict wind speed deficits and turbine power deficits that were compared
A simplified multiple wake model is developed by considering the wake of only the
nearest upstream turbine and its operational condition (wind speed, thrust coefficient and
• Wind turbine operating conditions and the extra flow momentum exchange
between multiple wakes and with the free stream will affect or contribute to
single turbine wake and interaction with the free stream which can be attributed
to the higher momentum exchange. Thus the wind speed in the ith overlapped
wake, 𝑢𝑖 , is
where 𝑈0 is the free stream velocity, 𝑢𝑖′ is the velocity calculated in the wake of the
upstream turbine by a single wake model, and C is a combination coefficient having the
36
When the upstream turbine is in the wake of another turbine, the wind entering the
rotor area has a lower wind velocity than the free stream. Thus, if momentum exchange
with the atmospheric boundary layer is ignored the wind velocity will at most only recover
to the wind speed at the upstream turbine. However, because of interactions with the
atmospheric boundary layer, it should eventually recover to the free stream velocity as the
To determine the value for the combination coefficient C, the contribution of the
free stream velocity was assessed as shown in Figure 3. The wake in this case was produced
by a wind turbine in the free stream with a thrust coefficient, 𝐶𝑇 , equal to 1. This means
that no free stream momentum flow could pass through the turbine. The wake velocity,
1
𝐶= 5.2
(1 + 2𝑘 ′ 𝑠)2
where 𝑘 ′ is the wake decay coefficient for this case. Its value should be less than 0.075
which is chosen for a normal wake, because there is a lower velocity gradient between the
inner wake and the free stream as shown in Figure 5.1. The wake decay constant, 𝑘 ′ , shown
in Figure 4, is a function of the ratio of upstream turbine velocity deficit to the free stream
velocity as following
𝑢𝑖−1
and 𝑘 ′ = 0.075 ≤ 0.8
𝑈0
37
Figure 5.3 shows the combination coefficient, C, calculated using equation (5.2) with
different 𝑘 ′ versus normalized downstream distance s. Figure 5.4 shows the wind speed
recovery calculated using the multiple wake model and the Jensen single wake model. The
wind velocity in the wake(s) caused by the wind turbine will finally recover to the free
stream velocity.
38
0.2
0.18
0.16
0.14
0.12
0.1
k'
0.08
0.06
0.04
0.02
0
0 0.2 0.4 0.6 0.8 1
The ratio of wind velocity at upstream turbine to free stream velocity
Figure 5.2 Wake decay constant, 𝒌′ , as a function of upstream turbine wind speed deficit
Figure 5.2 shows the relationship between coefficient 𝑘 ′ and wind speed at upstream
turbine. When upstream turbine is in free stream, 𝑘 ′ is zero and will increase with
decreasing wind speed ratio to freestream which means higher wake effects on the turbine.
39
0.8
0.7
0.6
0.5
0.4
C
k'=0.030
0.3 k'=0.053
k'=0.075
0.2
0.1
0
0 10 20 30 40
Normalized downstream distance
Figure 5.3 shows C calculated by equation (5.2), relating to upstream turbine wind
speed deficit and downstream distance. C is zero when downstream distance is zero,
because there is no space where the wake can interact with free stream. With downstream
distance increasing, C is increasing as the wake accumulates more free stream kinetic
energy. In addition, if the upstream turbine is under higher wake effects, it can be
considered to produce higher turbulence intensity. Then the increasing rate of C by the
downstream distance will be higher because higher turbulence intensity means more
energy exchanges.
40
0.9
0.8
0.7
Normalized wind velocity
0.6
0.5
1
Turbine in freestream
0.9 u1/Uo=0.81
u2/Uo=0.85
0.8
u3/Uo=0.9
0.7
0.6
0.5
0 10 20 30 40
Normalized downstream distance
Figure 5.4 Normalized wind velocity calculated using multiple wake model and the Jensen wake
single model. Blue: turbine in the free stream. Red: Second turbine with an incident wind velocity
of 0.81 of the free stream. Green: Second turbine with an incident wind velocity of 0.85 of the free
stream. Purple: Second turbine with an incident wind velocity of 0.9 of the free stream.
Figure 5.4 shows wind speed recovery calculated by our new multiple wake model.
Wind speed recovers slower than turbine wake in the free stream especially in the near
wake region. However, with downstream distance increasing (more than 3 to 4 D), the
recovery rate is similar to that from single wake model and will finally recover to free
41
5.1.2 Effect of the wind array on the atmospheric boundary layer
Investigations by Frandsen [15, 66] show that standard wake models under predict
wake effects. He states that the reason is that the effect a large wind farm has on the
atmospheric boundary layer is not taken into account. This effect was modeled as an
extended wind farm and resembles an increase in local surface roughness which results in a
change in vertical wind profile. According to this theory, the infinite wind farm equivalent
⎧ ⎫
𝜅
𝑧00 = ℎ𝐻 exp − 5.3
⎨ ⎬
�𝑐𝑡 + (𝜅/ln(ℎ𝐻/𝑧0 ))2
⎩ ⎭
where ℎ𝐻 is the hub height, 𝜅 is the von Karman constant, 𝑧0 is the local roughness without
𝜋
𝑐𝑡 = 𝐶 5.4
8𝑠𝑑 𝑠𝑐 𝑇
where 𝐶𝑇 is turbine thrust coefficient and 𝑠𝑑 and 𝑠𝑐 are the mean downwind and crosswind
spacing in rotor diameters. The wind farm roughness of 0.56 will be calculated for the Horns
As the wind reaches the wind farm, an internal boundary layer (IBL) is modeled by
Sempreviva et al. [72] using an increase in roughness. The effect on the wind speed at the
hub height for the turbine within the array is estimated from meteorological theory under
the assumption that the upstream wind flow and the IBL flow are neutral logarithmic
42
profiles. In the model the free stream speed, 𝑈0 , for a known height, z, above the internal
𝑢∗0 𝑧
𝑈0 = 𝑙𝑛 � � 5.5
𝜅 𝑧0
𝑢∗1 𝑧
𝑈1 = 𝑙𝑛 � � 5.6
𝜅 𝑧00
where 𝑢∗0 and 𝑢∗1 are the friction velocity for the free stream above and below the internal
boundary layer.
𝑈1 𝑢∗1 ln(𝑧/𝑧00 )
= ∙ 5.7
𝑈0 𝑢∗0 ln(𝑧/𝑧0 )
At the height h the wind speed above and below the internal boundary layer are the
same, hence
𝑢∗0 ln(ℎ/𝑧00 )
= 5.8
𝑢∗1 ln(ℎ/𝑧0 )
It follows then that when the height of the IBL grows higher than turbine hub height,
the new wind speed 𝑢 at turbine hub height can be calculated from equations (5.7) and (5.8)
and is
Frandsen et al. [15] developed a model of the growth of the internal boundary layer
𝜕ℎ 𝑐𝑚𝑤 𝑐𝑚𝑤
= 𝑐 → ℎ= 𝑐 (𝑥 − 𝑥𝑜 ) + ℎ0 5.10
𝜕𝑥 1 − 𝑐𝑚𝑤 𝑡 1 − 𝑐𝑚𝑤 𝑡
43
where 𝑐𝑚𝑤 is the relative flow speed in the wake, x is the downstream distance in the wind
farm and 𝑥𝑜 and ℎ0 are to be determined. Frandsen et al. [15] suggested that the value of
𝑐𝑚𝑤 could first approximated by the flow speed deficit in the infinitely large wind farm, 𝑐𝑤𝑓 .
Kristensen et al. [73] showed a relationship between the friction velocity before and after
Hence the flow speed deficit in the infinitely wind farm, 𝑐𝑤𝑓 , can be calculated by
However, Frandsen’s model on the growth of internal boundary layer was developed
from a wind direction that is along turbine rows. In order to account for other orientations
Based on observed data at Horns Rev [8, 60] the following assumptions were made
• Offshore roughness height 𝑧0 is equal to 0.0001 without the wind farm [66].
the height of the internal boundary layer reaches turbine hub height. This distance is
• Frandsen’s model on the growth rate of the internal boundary layer for the
wind parallel to wind turbine rows is extended for all other wind directions.
44
Based on these assumptions, the following empirical formula was developed for the
𝜕ℎ 𝑐𝑤𝑓 𝑐𝑤𝑓
= 𝑐 → ℎ(𝑥) = 𝑐 (𝑥 − 𝐷𝑠 ) + ℎ𝐻 5.13
𝜕𝑥 1 − 𝑐𝑤𝑓 𝑡 1 − 𝑐𝑤𝑓 𝑡
Figure 5.5 shows the calculated free stream speed at turbine hub height for the
1
Normalized freestream speed at turbine height
0.99
0.98
0.97
0.96
0.95
0.94
0.93
0.92
0 1000 2000 3000 4000 5000
Distance (m)
Figure 5.5 Predicted free stream wind speeds at turbine height for the Horns Rev wind farm layout
45
5.2 Wind array layout optimization model (WALOM)
In this section, the wind array layout optimization model is described. To distinguish
it from other models it is referred to as the WALOM or the WALOM model. It uses the LWAP
model to evaluate the turbine and wind farm power output. However, the multiple wake
model is simplified and effect on the wind profile in atmospheric boundary by the wind
farm is not considered in WALOM. Simulations of wind distribution, wind array layout and
turbine properties such as power curve were demonstrated. The Genetic Algorithm method
is also introduced and applied in the Matlab in which turbines’ positioning is optimized,
evaluated and analyzed under given simulated various wind farm circumstance.
The area occupied by the wind array is generally taken to be a rectangular area with
Cartesian coordinate system, see Figure 5.6. No matter whatever wind array site boundary
is, we can always find an appropriate rectangular to include all turbines’ locations. An x
coordinate and a y coordinate are given to each turbine to describe its location. Note that
all coordinates must be positive in our model for codes calculating wake effects.
For any given wind array, all wind turbine coordinates are then determined and
normalized ranging from 0 to 1. For example for Horns Rev array, turbine coordinates are
shown in Figure 5.7. In this way, we can create a general code which is suitable for wake
46
When evaluating wake effects, we first give each turbine a ranking arranged by
upwind location to downwind location order. For examples, when wind is from the north
direction, this ranking is ascending from north location turbine to south location turbine.
However, in East wind case, the ranking starts at east location turbine and ends at most
47
Figure 5.7 Horns Rev array turbine layout coordinates
48
Figure 5.9 Horns Rev turbines ranking for West wind
straight forward under certain controlled laboratory conditions, i.e., scaled wind turbine
models in wind tunnels or small turbines in large wind tunnels by Larwood [74], is more
complicated in the field because wakes are spatio-temporally variable phenomena shifting
with the direction of the wind and not amenable to meteorological measurements on a long
meteorological towers offshore is costly especially offshore and so wake measurements are
typically limited to measurements downstream of the prevailing wind direction. The Danish
49
wind turbine group at RISO has made the most advances in wake measurements. Three
meteorological towers erected in the vicinity of eleven turbines at the Vindeby offshore
farm have permitted simultaneous measurements of wind speed in the free stream and
wake for several wind directions by Frandsen et al. [20]. Measurements of velocity profiles
at different distances downstream from the turbines were also obtained by Barthelmie et al.
[23], using ship-borne Sonic Detection and Ranging (SODAR). Wake measurements are also
discussed by Barthelmie et al., in a recent paper [8]. Wake measurements have also been
obtained from satellite synthetic aperture radar (SAR) measurements by Christiansen et al.
[22].
In planning wind farms, shot-time wind data plays an important role in estimating
various engineering parameters, such as wake effect, power output, extreme wind load and
fatigue load. Raw data (including wind speed, wind direction etc.) will be tested for
validation and then be analyzed. Recording an average wind speed and wind direction of 10
minutes period is widely used. Table 5.1 is a recording sample 10 minutes average wind
This first wind data input method is to direct applying wind data into our program.
However, tens of thousands data are created even only for one year period wind recording.
Evaluating wake effects with this large number of wind cases need a lot of CPU time
50
Table 5.1 Wind data example
51
5.2.2.2 Weibull wind distribution
The wind variation for a site can be statistically described using a Weibull
distribution [76]. Wind speed data in our study is discretized by 1m/s. Wind orientation
where the direction-dependent Weibull parameters are 𝐾, the shape factor, 𝐴𝑐 the scale
factor.
Figure 5.10 Mean Wind Speed Frequency Distribution (Weibull shape factor, K, and scale factor,
𝐀 𝒄 , was 2.10 and 8.33 respectively)
52
Figure 5.10 shows an example of the Weibull wind distribution. The wind data is
from a 50m height anemometer. It shows a maximum frequency of 5.57% for the bin 6.5 to
7.0 m/s (14.5 to 15.7 mph). This is the modal point of the distribution. The average was
reported earlier to fall within the bin 7.0 to 7.5 m/s (15.7 to 16.8 mph) which occurs at a
frequency of 5.37%.
53
Figure 5.11 is a wind direction distribution rose for wind data from Lake Erie.
Prevailing winds come primarily from around southwest at nearly 45%. Southwest is the
most prevalent sector of the 16 with a frequency of 10.49%. Like the overall wind speed
frequency distribution, a direction wind rose cannot tell the whole story as about the wind
speed distribution. For each wind direction, Weibull factors and probabilities are calculated
Sector 𝐀𝒄 K %
54
5.2.3 Cost function
The cost function is needed when optimizing wind array layout problem for a given
site information while number of wind turbine is also not determined. A cost function
related to number of wind turbines was chosen that was also used in previous studies [27-
31]. The total cost is only dependent on the number of wind turbines, N, installed in the
2 1 2
𝑐𝑜𝑠𝑡 = 𝑁( + 𝑒 −0.00174𝑁 ) 5.15
3 3
The cost function is based on that a maximum discount of 1/3 is available when large
number wind turbines are purchased. Figure 5.12 shows the total cost of wind farm based
on number of turbines.
55
5.2.4 Turbine power
Wind turbine extract kinetic energy from the wind. According to wind turbine
momentum theory, the available power from the wind crossing a wind turbine for a wind
speed, 𝑢, is
1
𝑃𝑎𝑣𝑎𝑖𝑙 = 𝜌𝐴𝑢3 5.16
2
𝐶𝑇 = 4𝑎(1 − 𝑎) 5.18
Figure 5.13 shows power curve and correlated thrust coefficient for turbines
installed on Horns Rev wind farm. Note that turbine cut in speed is 4m/s and cut out speed
is 25m/s.
atmospheric boundary layer by existing wind array will be evaluated and an affected new
free stream field at the turbine hub height will be calculated. Then wind turbines are
checked in the upstream to downstream order that whether they are operating in the wake
of any other wind turbine. If this is not the case, then power is calculated using the
atmospheric boundary layer considered free stream velocity. Otherwise wind velocity at the
point where the wind turbine placed is determined by applying wake model discussed
earlier.
56
Figure 5.13 Vestas V80 2MW turbine power output and thrust coefficient vs. wind speed [78]
In WALOM model, the LWAP model is used to evaluate wake effects and predict turbine
power output and the whole wind farm performance with a given array configuration. However the
LWAP model used in WALOM was modified: (1) the multiple wake model was simplified that only
the wake by nearest upstream turbine were calculated; (2) the effect on the wind profile in
57
5.2.6 Genetic Algorithm (GA)
The so called genetic algorithm has been a popular approach so far although it is not
the only option. The genetic algorithm is a stochastic global search method that evolves
“chromosomal” string that represents say, two individuals, and is used to create the genetic
code of a new individual with its own code. During the reproduction phase, each individual
is assigned a fitness value derived from its raw performance measure given by the objective
function. This value is used in the selection to bias towards more fit individuals. Genetic
operators manipulate the “genes” of the chromosomes between parent pairs by crossover
and mutation, producing child generation. After recombination and mutation, the individual
strings of new generation are then decoded, assigned with fitness values, and then
compared. The process continues through subsequent generation. Figure 5.14 shows a
A wind farm layout evaluation code is developed in MATLAB which calculates the
power produced and the cost of a wind farm. The code is coupled with the genetic
algorithm solver (referred as ‘ga’ solver), available in MATLAB’s genetic algorithm toolbox
for optimization process. In our code, wind turbine position coordinates was developed. The
number of turbines N must be fixed first, and there are 2N variables represent X and Y
coordinates. In this algorithm, turbine can be placed any location compare to center of cells
“chromosomal” strings that represent individuals, and is used to create the genetic code of
a new individual with their own codes. For given number of turbines, the placement of
58
these turbines is described in the genetic string. Evolution and adaptation of the individual
strings representing turbines’ placement will be applied and guarantee “best placement
individuals” with higher power output for a given wind distribution. In GA judgment of
better turbine placement is by the objective function, also referred to as the fitness value
𝑐𝑜𝑠𝑡
𝑂𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 = 5.19
𝑃𝑡𝑜𝑡𝑎𝑙
The flowchart in Figure 5.15 demonstrates the process through of the code and the
‘ga’ solver operates to find an optimal solution. The optimization process starts with the
initialization in the genetic algorithm ‘ga’ solver. In the initialization step, following
turbines because two variables are required to specify the position of a wind
• Population size: The population size is the total number of individual solutions
• Constraints: The constraints in the ‘ga’ solver are specified as bounds i.e., lower
and upper limits for the variables. The area size of the wind farm is specified in
constraints so that wind turbines cannot be placed outside the wind farm region.
criteria in the ‘ga’ solver. It include maximum number of iterations which is also
59
function value over stall generations is less than function tolerance than
After the initialization process, random set of solutions is created taking into account
the constraints. All the solutions created are analyzed by the wind farm layout evaluation
model. Estimated power production and the cost of the wind farm are calculated and
objective function value (cost per unit power) of each solution is returned to ‘ga’ solver.
In the next step, the optimization criteria are checked if they are satisfied or not.
When the optimization criteria are not met, all the solutions are ranked based on their
objective function values. A solution with small objective function value is better as its cost
per unit power is smaller and is placed before other solutions with larger objective function
value.
After ranking is completed, a number of solutions are selected and some new
solutions are created (reproduced). This selection of solutions is affected by the ranking
done in previous step and a solution with good ranking has a better chance of being
selected. New solutions are created while some solutions are copied from original set of
solutions to the new set of solutions. These selected a few solutions are one of the best in
The last step before new set of generations (new population) is ready is called
Mutation. In this step some random changes are made in a few solutions. This step is very
important as it helps in maintaining diversity in the solution set. This new solution set is
analyzed by the wind farm performance evaluation code and this iterative procedure
60
With application of a large number of evolutionary iterations, the sequential
placement configurations of turbines are iterated toward an optimal. Figure 5.15 shows a
61
Figure 5.15 Flowchart of Genetic Algorithm optimization process
62
Chapter 6 Results: wind farm layout evaluation model
Matlab codes for the LWAP model were developed to apply the Jensen wake and
Ainslie wake models to predict wake losses at the Horns Rev wind array and Nysted wind
arrays. Wind array configurations for wind farm and site information, wind distribution and
offshore conditions are taken into account. Power curves and parameters corresponding to
the Vestas V80-2.0 MW turbines located in Horns Rev and the Siemens 2.3 MW turbines
Results of wind speed deficits and turbine power deficits from multiple wind cases
are evaluated and compared to observation data by Rathmann et al. [59] and Barthelmie et
al. [8, 60]. Evaluations made using the model developed here are also compared to other
wind farm analyzing tools such as WAsP, WindFarmer and NTUA etc. Compared to other
models, our model applies simple computational schemes. For example, with a given wind
efficiency will be processed in 2 seconds using the Jensen wake model and 10 seconds with
6.1 Wind speed evaluations at Horns Rev when wind direction is along
turbine rows
In this case, two wind directions along wind turbine rows in the Horns Rev Array are
taken into consideration, shown in Figure 6.1. Wind turbine spacing is 7D in the 270° case
and 9.4D in the 222° case. Wind speed deficits at turbine locations in the row are evaluated
with two free stream speeds (8.5 m/s and 12m/s) and compared to observation data by
63
Rathmann et al. [60]. There are two groups of observation data, respectively from internal
row located in the interior of the wind farm and external row located at the edge of wind
farm. At these low to moderate wind speeds, the thrust coefficient is relatively high, and
Figure 6.1 Horns Rev layout: Case 1 of 270° and 7D spacing, Case 2 of 222° and 9.4 D spacing
64
1
Observation
0.95 Jensen model
Ainslie model
Wind velocity deficit
0.9
0.85
0.8
0.75
0.7
0 2 4 6 8 10 12
Turbine number
Figure 6.2 Horns Rev evaluation wind direction 270° and wind speed 8.5 m/s +/- 0.5 m/s
In Figure 6.2, evaluation of wind speed deficits at each turbine location from wind
direction of 270° and wind speed of 8.5 m/s is obtained and compared to observation data.
Results show that evaluation by Ainslie model in general under predicted wind speed deficit
by about 0.01 to 0.02 of free stream velocity. The Jensen model seems to have a better
65
1
Observed int. row
Observed ext. row
Jensen model
0.95
Ainslie model
Wind velocity deficit
0.9
0.85
0.8
0.75
0 2 4 6 8 10 12
Turbine number
Figure 6.3 Horns Rev evaluation wind direction 270° and wind speed 12 m/s +/- 0.5 m/s
In Figure 6.3, evaluation and observation data of wind speed deficits are from wind
case of 270° and 12m/s. Observations are from both internal turbine row and external
turbine row. Wind speed deficits at internal turbine are less than those at external turbine.
This may be because there is a more complex wake effect well inside the array, so that
turbulence intensity will be higher. This means that more energy is exchanged between the
turbine wakes and free stream wind. Clearly, the Jensen and Ainslie models generally over
predict the wind speed deficit in the wake. Predictions are better correlated to observations
66
from l turbine rows on the outside of the array. The Ainslie model appears to yield better
0.95
Wind velocity deficit
0.9
0.85
Figure 6.4 Horns Rev evaluation wind direction 222° and wind speed 8.5 m/s +/- 0.5 m/s
In Figure 6.4, wind speed deficits are evaluated with a wind direction of 222° and
wind speed of 8.5 m/s. As discussed in Figure 6.3, observed wind speed deficits from
interior turbine rows are less than those from exterior rows. This may also because of
higher turbulence intensity inside the wind farm. Predictions by our model are not as good
as in the 270° wind direction case. Wind speed deficits are over predicted for the first three
67
1
0.95
0.9
Wind velocity deficit
0.85
0.8
0.65
0 2 4 6 8 10
Turbine number
Figure 6.5 Horns Rev evaluation wind direction 222° and wind speed 12 m/s +/- 0.5 m/s
In Figure 6.5, it shows that predictions for wind speed of 12m/s on 222° wind
the mean absolute percentage error (MAPE) of the wind velocity was calculated using
𝑛
100% 𝑢𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 − 𝑢𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑
MAPE = �� � 6.1
𝑛 𝑢𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑
𝑖=1
where u is the wind velocity at each turbine (observed or predicted) and n is the number of
turbines.
68
The MAPE for the predictions by the LWAP using the Jensen wake model and the
Ainslie wake model versus observation data of mean wind speed are calculated in Table 6.1.
It shows that observations by the LWAP using both single wake models have MAPE less than
2.2%. Predictions from exterior turbines are better correlated to observation data (i.e.,
exhibit lower MAPE) than interior turbines. In addition, predictions for 270° wind direction
Table 6.1 MAPE of the computed normalized wind velocity using observed data for two
wind directions and two wind speeds reported in [60] at Horns Rev.
Wind turbine power can be predicted by evaluating the wind velocity at all turbine
locations within the array and then calculating the predicted output power using a wind
turbine power curve [32]. The curve shown in Figure 6.6 illustrates the one used in the
model and represents a Vestas-V80 2 MW wind turbine in Horns Rev. Two wind direction
cases (the 270° case and 7D spacing, the 222° case and 9.4D spacing) along wind turbine
69
6
x 10
2.5
1.5
0.5
0
0 5 10 15 20 25 30
Wind velocity (m/s)
0.95
0.9
Observation
0.85 Jensen model
Ainslie model
Normalized power
0.8
0.75
0.7
0.65
0.6
0.55
0.5
0 1 2 3 4 5 6 7 8 9
Turbine number
Figure 6.7 Turbines power at Case 1 for wind speed at 8m/s and direction 270° at Horns Rev
70
1
0.95
0.9
Observation
0.85 Jensen model
Normalized power
Ainslie model
0.8
0.75
0.7
0.65
0.6
0.55
0.5
0 1 2 3 4 5 6 7 8 9
Turbine number
Figure 6.8 Turbines power at Case 1 for wind speed at 10m/s and direction 270° at Horns Rev
0.95
0.9 Observation
Jensen model
Normalized power
0.8
0.75
0.7
0.65
0.6
0.55
0 1 2 3 4 5 6 7 8
Turbine number
Figure 6.9 Turbines power at Case 2 for wind speed at 8m/s and direction 222° at Horns Rev
71
Figures 6.7 and 6.8 show results of wind turbine power calculated for a wind
direction of 270° and wind speeds of 8 m/s and 10 m/s. The second case is the wind
direction of 222° and a wind speed of 8 m/s (chosen because there is a real data set to
compare with). Results are shown in Figure 6.9. In this case, Barthelmie et al. [8] reported
data of the first 5 turbines because there was a large uncertainty in the measurements due
The mean absolute percentage error of the normalized turbine power was
where P is the normalized power to turbine power at the free stream at each turbine
The MAPE for the turbine power are shown in Table 2. Predictions using the Ainslie
single wake model have a higher MAPE in 270° wind direction cases than those using Jensen
Table 6.2 MAPE of the computed normalized turbine power using computed predictions
and actual observations at Horns Rev [58]: wind direction parallel to turbine rows.
72
6.3 Power output predictions for turbines in the row at the Horns Rev
and Nysted for a representative wind speed and variable wind
directions
Detailed case studies of turbine power losses due to wakes at the Horns Rev and
Nysted wind farms were analyzed. The major difference between the two wind farms is the
turbine spacing, with 7 × 7D at Horns Rev and 5.8 × 10.5D at the Nysted. The average power
at each turbine was calculated for seven wind directions: a wind direction where the flow is
down an exact row (ER) including observations within ± 2.5° (270° ± 2.5° at Horns Rev, 278°
± 2.5° at Nysted), and six directions of ER + 5°, + 10°, + 15°, - 5°, - 10° and - 15°, as shown in
Figures 6.10 and 6.11. The observed data were reported by Barthelmie et al. [60] and
results by wind farm layout evaluation models such as WAsP, WindFarm, WakeFarm and
NTUA were also reported and analyzed. As shown in Figure 6.12 and 6.13, results from
Horns Rev and Nysted illustrated an acceptable agreement between predictions and
observations in most directions except 255° in Horns Rev and ER ± 5° in both Horns Rev and
Nysted arrays. It was reported that in 255° wind direction at Horns Rev the asymmetry in
the observations may not reflect real case and it could be a data issue due to insufficient
observations [60].
73
Figure 6.10 Horns Rev Array. Exact Row (ER=270°) of turbines [60]
74
Figure 6.12 Normalized power at Horns Rev for the free stream wind speed of 8 ± 0.5 m/s:
comparison of models with observations
75
Figure 6.13 Normalized power at Nysted for the free stream wind speed of 8 ± 0.5 m/s:
comparison of models with observations
76
In order to compare predictions by the model to reported results of other wind farm
layout evaluation models, the root mean square deviation (RMSD) of the normalized
2
∑𝑛𝑖=1�𝑃𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑 − 𝑃𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 )�
RMSD = � 6.3
𝑛
The RMSD of the turbine power predictions by models versus observation data of
the mean turbine power by various wind directions and wind speed of 8 m/s at Horns Rev
and Nysted wind farms are given in Tables 3 and 4. In general, predictions in by the LWAP
using either the Jensen and Ainslie single wake model have the same accuracy, with the
overall RMSE of 0.08 by using the Jensen model compare to 0.07 by the Ainslie model in
Horns Rev case. The overall RMSE for Nysted wind farm are 0.05 computed using the Jensen
model and 0.04 b using the Ainslie model. However, results computed using the Jensen
model show that there are large discrepancies between predictions and observations at
Horns Rev for a 265° and 275° wind direction cases and also for the Nysted 283° wind
direction case. This is likely due to the uniform wind velocity distribution in Jensen wake
model which tends to desensitize the sensitivity of this model to wind direction.
Compare to the RMSD by other wind farm layout evaluation models, the LWAP
performs better (i.e., exhibit lower RMSD) than WAsP. These predictions also have little
RMSD difference compare to CFD based models such as WindFarmer, WakeFarm and NTUA.
However, the run times for the LWAP applied to a full wind farm simulation like Horns Rev
wind farm is only a few seconds. This is significant less than reported time for running in the
77
Table 6.3 RMSD of the computed normalized power for various wind directions and a
wind speed of 8 m/s at Horns Rev wind farm [60]
Table 6.4 RMSD of computed normalized power for various wind directions and a wind
speed of 8 m/s at Nysted wind farm [60]
78
Chapter 7 Results: wind farm layout optimization
In this chapter, the wind array layout optimization model (WALOM) is applied with
Genetic Algorithm method to the wind farm layout optimization problem. Wind turbine
layout is optimized by a given site information and wind distribution, while wake effects are
work started by Mossetti et al. [25], in which a 2km × 2km square wind farm site area is
considered. Mossetti also assumed a simple wind distribution with a higher frequency for
some directions and applied the Jensen wake model in the calculation. In this work the
Mossetti approach was extended by developing a new turbine placement coordinates and
considering effect on the results by factors as turbine spacing limit, area size and wake
Section 7.2 describes the second approach which is from the view point of the
existing or potential wind farm layout design. Unlike the approach discussed in section 7.1,
the wind data, site area, turbine properties etc., from a real wind farm site were used.
Weibull distributions were applied and turbulence intensities of wind were also factored
into the model. A real wind turbine power curve was used to convert wind speeds to power.
Three cases for an area of 2 km ×2 km square were investigated from section 7.1.1
to section 7.1.3. Case 1 is a simple example for a uniform wind direction with a free stream
velocity of 12 m/s, shows in Figure 7.1. There are two reasons for looking at this, one is to
79
compare and validate against previous work and two, understand the effects of parameter
variation in the model. For Case 2 the wind direction is variable and the free stream velocity
is constant at 12 m/s. There is an equal probability that wind blows from any direction. The
wind direction is discretized in 36 segments each measuring 10°. Case 3 is variable wind
direction and variable wind speed case. Figure 7.2 shows the wind distribution in this case.
Three wind speeds are possible, 17, 12 and 8 m/s. The probability is higher for wind
The optimization for wind Case 1 (uniform one speed wind) was also applied for
different wind turbine spacing limits (the smallest distance between two turbines) and area
sizes (2 km2, 4 km2, 8 km2 and 16 km2). Results are shown in sections 7.1.4 and 7.1.5.
In section 7.1.6, the optimization was applied to wind Case 2 (equal probability
variable wind direction with a constant wind speed) with a round shape wind farm site area.
The reason for studying on this is that it is expected to find some optimized layout pattern
for the turbine layout under isotropic wind distribution and isotropic site shape area.
In section 7.1.7, optimized turbines layouts using the Jensen and Ainslie wake
models were calculated and compared for the model three wind cases presented above.
The optimization was carried out for different numbers of turbines (N). For each N,
there is an optimal configuration. The space limit between two turbines is such that two
turbines should not occupy the same 100m × 100m square. Figure 7.3 shows optimal fitness
values for different N. The limiting value of fitness is the situation when there is no wake
80
effect for all turbines so that a maximum power is produced. It is noted that when N
increases, the difference between the optimal fitness value and the limiting value increases.
The optimal result is when N=48 and is compared with previous studies in Table 7. 1.
Figure 7.4 shows the computed optimal distributions of wind turbines from previous
studies and this study. All were configured under the same parameters except that in the
present study two turbines cannot occupy the same 100m × 100m square. The grid is
sufficiently refined following Mittal’s approach [29] that turbines can be placed anywhere
In the Figure 7.4 (E), it shows that turbines trend to be positioned at upstream area
and downstream area. This is because in upstream area, there is no wake affect and
downstream area has better wake recovery. This is unlike the optimized layouts by other
works, shown in Figure 7.4. It may because of a different multiple wake model used in
WALOM. It is also noted that placement of diamond shape is preferred in the distribution as
it reduces the possibility that wind speeds will be lowered at the turbine because of the
Table 7.1 compares the number of turbines, the total power produced, fitness value
and efficiency for the results of previous studies and the present study. The optimal
placements of previous studies were also recomputed by our model. The power outputs of
81
Figure 7.1 Wind farm area
82
Figure 7.3 Fitness value of different number of turbines for case 1.
Table 7.1 Results from previous study and current study: reported and recomputed
83
A B
↓ ↓
C D E
↓ ↓ ↓
Figure 7.4 Turbines placement of four studies for case 1: (a) Mossetti et al. [25] (b) Grady et al.
[27] (c) Marmidis et al. [28] (d) Mittal et al. [29] (e) WALOM
84
7.1.2 Case 2: Constant wind speed with an equal probability variable wind direction
In this case, several optimal layout configurations were obtained. Figure 7.5 shows
the fitness values for different number of turbines. When N=32, the fitness value reduces to
a minimum of 1.5749×10-3. The optimized placement of the turbines for this case and those
85
A B
C D
Figure 7.6 Turbines placement for case 2: (a) Mossetti et al. [25] (b) Grady et al. [27] (c) Mittal et
al. [29] (d) WALOM
Table 7.2 is a comparison for the results of previous studies and the present study.
Even though the optimal configuration obtained does not have highest efficiency the fitness
value is lowest.
Table 7.2 Results from previous study and current study: reported and recomputed
86
7.1.3 Case 3: Variable Wind Speed with Variable Wind Direction
In this case, the optimal configuration was obtained when N=35. Figure 7.7 shows
optimal fitness values for different number of turbines. The optimal wind farm layout
configuration of previous studies and that obtained using the model described earlier are
optimal wind turbine location will tend to be positioned in the edges of the turbine array
area. Presumably this is an attempt to maximize the distance between turbines and to
87
eliminate the effects of wakes, like what had been discussed in section 7.1.1. It is worth
noting that, because the proximity exclusion rule has been relaxed in comparison to the
restrictions in previous models which did not allow turbines to occupy the same 200m ×
200m square, turbines at the edges cluster in a zigzag fashion along the edge direction. This
is to reduce wake effects when the wind blows along the edge of the square.
Table 7.3 compares the results obtained in previous studies and this study. It shows
that the efficiencies recomputed of previous configurations were reduced. The optimal
A B
C D
Figure 7.8 Turbines placement for case 3: (a) Mossetti et al. [25] (b) Grady et al. [27] (c) Mittal et
al. [29] (d) WALOM
88
Table 7.3 Results from previous study and current study: reported and recomputed
In this section, results were obtained by setting spacing limits of 50m × 50m, 100m ×
100m, 150m × 150m and 200m × 200m, for the case of unidirectional wind of the same 2km
by 2km square area. Four groups of data about fitness value and efficiency are compared.
Figure 7.9 shows the fitness values changes depending on the number of turbines and
Figure 7.10 shows the fitness values depending on the number of turbines for each spacing
limit.
The fitness value increases and the efficiency decreases faster for large spacing limit,
89
Figure 7.9 Fitness values of different spacing limits
90
50m × 50m 100m × 100m
Optimal placements of 40 turbines for the four spacing limits are plotted in Figure
7.11. More wind turbines are positioned in the center area of the square for large spacing
limits. It is probably because large spacing limit provides less placement options in upstream
91
7.1.5 Different Area Sizes for Unidirectional Wind
Optimization for area sizes of 2 km2, 4 km2, 8 km2 and 16 km2 were applied. The
spacing limit is 100 m ×100 m in the case. Figures 7.12 and 7.13 show the fitness values and
efficiencies of for each area size. Results indicate that the velocity deficit is much larger for
an area of 2 km2, but 35 turbines can still have a total efficiency of 95%. With increasing
Figure 7.14 shows the optimal placement of 40 turbines for each area size. It is
noted that for all cases, the optimal positions of turbines tend to cluster on upstream and
downstream edges. This is consistent with the discussion for placement of wind turbine for
Case 1. It is also found that for the square area of 2 km2, the array efficiency drops rapidly
92
Figure 7.12 Fitness values of different area sizes
93
2 km2 4 km2
8 km2 16 km2
94
7.1.6 Constant Wind Speed with Variable Wind Direction and a circular site area
In this case, the wind distribution was the same as Case 2 described in section 7.1.2.
However, the optimization was applied to a circular wind farm area with a diameter of 2 km.
Figure 7.15 shows the fitness values for different number of turbines for this case. The
-3
x 10
2
1.8
1.7
Fitness value
1.6
1.5
1.4
1.3
1.2
0 5 10 15 20 25 30 35 40 45 50
Nuber of turbines
95
20 turbines 30 turbines
2000 2000
1500 1500
1000 1000
500 500
0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
40 turbines 50 turbines
2000 2000
1500 1500
1000 1000
500 500
0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
Table 7.4 Turbines distributions in optimized layouts for isotropic wind and round area
Number of turbines Near area boundary Inside the area % of turbines near the
boundary
20 15 5 75
30 21 9 70
40 27 13 68
50 32 18 64
96
Figure 7.17 shows that wind turbines tend to be optimally positioned at the
boundary of the wind array area. This is similar to results obtained for variable wind
direction and square wind arrays presented earlier. Table 7.4 shows the number of turbines
near the site boundary versus the number of turbines inside the area. It shows that optimal
turbine locations are located at the site boundaries. With an increasing number of turbines,
more turbines will be optimally positioned within the interior area of the site. This can be
explained simply as an attempt to maximize the distance between turbines and thus
7.1.7 Comparison of optimized layouts using Jensen wake model and Ainslie wake
model
In this section, optimized turbines layouts using Jensen wake model and Ainslie wake
model were obtained for 48 turbines with uniform wind (Case 1) and 32 turbines with
variable wind direction (Case 2) and a square area. Results are shown in Figures 7.18 and
7.19.
Figure 7.18 shows that optimal locations obtained using the Ainslie wake model
differ from those obtained using the Jensen wake model and tend to be located near the
upstream area and downstream boundaries of the array area. This is likely due to
assumption of a Gaussian wind velocity profile in the Ainslie wake model. In comparison to
the uniform wind velocity used in Jensen wake model, if a turbine is located at the edge of a
wake near another turbine, the wake deficit would be much less for the Ainslie model. Thus
the ‘proximity constraint’ that requires some distance between neighboring turbines due to
97
wake effects is relaxed somewhat for the Ainslie model. . In addition, turbine spacing in the
Figures 7.19 and 7.20 show optimized layouts for Case 2 and Case 3 wind
distributions using both wake models. A major difference between optimized layouts
obtained using the Ainslie wake model to those obtained using the Jensen wake model is
that turbines show less of a tendency to be positioned near the boundary of the area. This
is especially evident in Case 3, where more turbine spacing in the prevailing wind direction
1800 1800
1600 1600
1400 1400
1200 1200
1000 1000
800 800
600 600
400 400
200 200
0 0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 7.18 Optimal turbines layouts for 48 turbines with Case 1 uniform one direction (from
North to South) wind using Jensen wake model and Ainslie wake model.
98
Jensen wake model Ainslie wake model
2000 2000
1800 1800
1600 1600
1400 1400
1200 1200
1000 1000
800 800
600 600
400 400
200 200
0 0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 7.19 Optimal turbines layouts for 48 turbines with Case 2 for variable wind direction and a
constant wind speed using Jensen wake model and Ainslie wake model.
1800 1800
1600 1600
1400 1400
1200 1200
1000 1000
800 800
600 600
400 400
200 200
0 0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 7.20 Optimal turbines layouts for 48 turbines with Case 3 for variable wind direction using
the Jensen wake Ainslie wake models.
99
7.2 Horns Rev wind farm layout optimization
is applied the spatial configurations of wind turbines an actual offshore wind arrays. This
performance of that wind array. For the study the power curve, thrust coefficients and size
characteristics of a Vestas V80 wind turbine (This is the actual wind turbine installed at the
Horns Rev)e employed. The power and thrust coefficients are given in Table 7.4 [78]. The
direction-dependent Weibull parameters 𝐾, the shape factor, A𝑐 the scale factor and
probabilities as a function of wind orientation sector are shown in Table 7.5 [77]. The
average turbulence intensity in the Horns Rev for different wind speeds was computed by
cases were considered: 12, 36 and 72 possible directions. Optimized turbines layout results
applying these three cases were obtained and are presented in sections 7.2.1 to 7.2.3. In
section 7.2.4, optimized results obtained using Jensen wake model are discussed and
100
Table 7.4 Vestas V80 thrust coefficient and power as a function of wins speed
101
Table 7.5 Wind distribution for Horns Rev
Sector A𝑐 K %
Mean 10.71 2.33 100.0
N 8.65 2.11 5.1
NNE 8.86 2.05 4.3
ENE 8.15 2.35 4.4
E 9.98 2.55 6.6
ESE 11.35 2.81 8.9
SSE 10.96 2.74 6.5
S 11.28 2.63 8.7
SSW 11.50 2.40 11.5
WSW 11.08 2.23 12.1
W 10.94 2.28 11.1
WNW 11.27 2.29 11.4
NNW 10.55 2.28 9.6
102
7.2.1 Weibull data for 12 wind direction sectors
investigated. The Weibull data is used as described above. Each sector is 30°. The direction
The optimized wind farm layout of 80 turbines is shown in Figure 7.22. The average
power output of the wind array is 80.4 MW with an improvement of 8.8% over the
simulated Horns Rev. The optimized spatial configuration captured an additional 57 GWh.
Figure 7.23 shows the total annual energy produced in GWh of all turbines for each
sector, for Horns Rev and the optimized configurations. It is noted that the most
improvement occurs when wind flows parallel to Horns Rev turbine array direction. This is
103
Figure 7.22 Optimized layout for case 1
104
7.2.2 Wind data for 16 wind direction sectors
distribution. The 12 sector resolution of the wind direction data may not be sufficient so the
consequences of adding more angular resolution into the description of the wind direction
configurations will arise for different angular discretizations, some test simulations were
carried out. An actual Horns Rev wind distribution for 16 sectors was prepared and input
into the simulation model. The angle between two sector sides is 22.5°. For the 16-sector
case, the wind speed probability was calculated for each sector using the average Weibull
parameter for the Horns Rev wind field. Figure 7.24 shows the probability distribution for a
16 sector field. Figure 7.25 shows optimized result for 12 direction sectors. Obviously, the
For most wind direction sectors, the actual rectangular array layout of Horn Rev has
better performance. In addition, turbines in the rectangular layout can obtain 641 GWh of
annual power, more than the 636 GWh calculated for the optimization layout using 16 wind
direction sectors. This demonstrates that the degree of resolution of the angular
discretization of the input wind data can significantly influence the optimization results.
105
Figure 7.24 Wind direction discretization for 16 direction sectors
106
7.2.3 Wind data for 72 wind direction sectors
To address the problem discussed above, more wind direction data with higher
data with 72 orientation sectors was applied. The optimal layout configuration for this wind
field was obtained and compared to the predicted annual power output for the Horns Rev
configuration subject to the same wind field. As expected, the optimized configuration
produces more power. The wind direction probability density is shown in Figure 7.26.
107
Figure 7.27 Optimized layout for 72 direction sectors
Figure 7.27 shows the optimized layout. For this idealized case with constant wind
speed, the improvement is of 5.4% over the computed power for the actual Horns Rev
layout but subject to the synthesized wind field– 668 GWh compare to 704 GWh. It appears
that, with more wind direction sectors, there are more instances where there is an
increased probability that a turbine lies within in the wake of proximal upstream turbines.
Note that, even for this idealized wind field, the improvement of efficiency by 5.4% is not
108
Table 7.7 Layout performcance of optimization results
In Table 7.7, Horns Rev Array and two optimized array configurations from two wind
direction discretization cases (12 wind directions and 72 wind directions) were evaluated by
both wind discretization methods. It shows that the optimized configuration with 12
direction discretization produces less power than predicted for either the Horns Rev
optimized configuration from the 72 discretization produces more power than Horns Rev in
109
Chapter 8 Conclusions and future work
evaluate wind speed deficit and turbine power for large wind farms. The Jensen wake
model and the Ainslie wake model are applied to predict wake losses for Horns Rev wind
array and Nysted wind array. Predictions of wind speed deficits and turbine power losses
are evaluated and compared to observations and also other wind farm analyzing tools such
as WAsP, WindFarmer and NTUA etc. Results show that the LWAP model gives very good
predictions on both wind speed deficit and turbine power output on multiple wind cases. In
general, our predictions with Ainslie wake model are better than those with Jensen wake
model. Compare to other models, our evaluation results are better than engineering model
WAsP and as good as CFD based models such as WindFarmer and NTUA. In addition, our
model applies a general and straightforward multiple wake mode and uses simple
computational schemes. Thus, LWAP computational times are much less than those CFD
based model and LWAP can be readily applied to different array configurations and wind
The LWAP model is also applied in WALOM with a Genetic Algorithm to analyze the
wind farm layout optimization problem. Wind turbine placement is optimized by a given site
information and wind distribution, so that the wake effects are minimized and therefore the
In the case studies of extension of previous approach, three different cases are
optimized and recomputed. It shows that in Case 1, 48 turbines can be placed with the
objective function value of 1.3164×10-3 and an efficiency of 98.056%. This is far better than
110
previous studies. In Case 2 and Case 3, 32 and 35 wind turbines obtain optimal objective
function values of 1.5749×10-3 and 8.3852×10-4. Both of them are better than previous
results. Results for Case 1 of different spacing limits and area sizes were obtained. These
two factors affect the optimal fitness value and, thus, the spatial configuration of wind
turbines in the array. It suggests that when designing a wind farm, consideration of site are
conditions (size and shape) turbine spatial proximity limits and wake models should be
made.
The second wind farm layout optimization approach by WALOM have been
proposed for real wind farm site such as Horns Rev and applies real site information and
wind distribution. Ainslie’s eddy viscosity wake model is employed. Wind data including the
Weibull distribution, turbulence intensities and wind turbine characteristics are used. The
effects of wind direction discretization sectors are studied to evaluate the reliability of
optimization results. However, the optimized results are highly affected by discretization of
the wind data. More improved resolution for wind discretization is, thus, recommended for
wind array optimization method. Results show that for realistic wind fields with variable
direction and speed distributions, the performance increases for the optimized array will
that the optimized results are highly affected by discretization of the wind data. More
improved resolution for wind discretization is, thus, recommended for wind array
optimization method.
The LWAP Model has shown a reasonable degree of agreement with the data sets
from Horns Rev and Nysted wind. Other wake models are recommended to be applied in
111
future work such as Larsen’s wake model and the Fuga wake model. It is also suggested that
multiple wakes model used in LWAP could be extended to consider wake effects that
It has been shown that factors such as wind turbine spacing limits, site area sizes and
wind data resolution can strongly affect the results of wind array performance and array
optimization calculations. Other factors such as complex terrain and wind turbine type will
In addition, current optimized wind turbine placements seem to be very random and
subject to a lot of uncertainty. As a result, more optimization results are needed and more
112
Bibliography
[1] Global wind report – annual market update 2010, GWEC 2011.
[2] Renewables 2011 global status report, Paris, REN21 Secretariat, 2011.
[3] Fthenakis V, Kim H C. Land use and electricity generation: A life-cycle analysis.
Renewable and Sustainable Energy Reviews 2009; 13: 1465.
[4] World Wind Energy Report 2010. World Wind Energy Association. February 2011.
Retrieved 8-August-2011.
[5] Sawin J L, Wind Power Increase in 2008 Exceeds 10-year Average Growth Rate.
Worldwatch.org. Retrieved 29 August 2010.
[6] BTM Forecasts 340-GW of Wind Energy by 2013. Renewableenergyworld.com. 27 March
2009. Retrieved 29 August 2010.
[7] BTM Consult. International Wind Energy Development World Market Update, 2009.
[8] Barthelmie RJ, Hansen K, Frandsen ST, Rathmann O, Schepers JG, Schlez W, Phillips J,
Rados K, Zervos A, Politis ES and Chaviaropoulos PK. Modelling and measuring flow and
wind turbine wakes in large wind farms offshore. Wind Energy 2009; 12: 431–444.
[9] Sanderse B, Aerodynamics of wind turbine wakes: literature review. ECN 2009. Report
ECN-E-09e016.
[10] Thomsen K, Sørensen P. Fatigue loads for wind turbines operating in wakes. Journal of
Wind Engineering and Industrial Aerodynamics 1999; 80: 121–136.
[11] Hasager CB, Rasmussen L, Peña A, Jensen LE, Réthoré PE. Wind Farm Wake: The Horns
Rev Photo Case. Energies. 2013; 6:696-716.
[12] Mechali M, Jensen L, Barthelmie R, Jensen LE, Réthoré PE. Wake effects at Horns Rev
and their influence on energy production. Wind Energy Conference and Exhibition 2006.
Athens, Greece: p. 10.
[13] Barthelmie, RJ and Jensen LE. Evaluation of wind farm efficiency and wind turbine wakes
at the Nysted offshore wind farm. Wind Energy, 2010; 13: 573-586.
[14] Patel MR. Wind and Power Solar Systems. Boca Raton: CRC Press, 1999.
[15] Ammara I, Leclerc C, Masson C. A viscous three-dimensional differential/actuator-disk
method for the aerodynamic analysis of wind farms. Journal of Solar Energy Engineering
113
2002; 124:345–56.
[16] Schlezt W and Neubert A. Special Wake Cases, presented at VindKraftNet: Wake
Meeting 16th September 2008 at Dong Energy, Copenhagen. (See also, Proceedings
DEWEK 2006 Bremen, November 2006 “New developments in precision wake
modelling”, Wolfgang Schlez, Anja Neubert, Gillian Smith).
[17] Frandsen S, Barthelmie R, Pryor S, Rathmann O, Larsen S, Højstrup J, Thøgersen M.
Analytical modelling of wind speed deficit in large offshore wind farms. Wind Energy
2006; 9: 39–53. DOI: 10.1002/we.189
[18] Frandsen S, Rathmann O, Barthelmie R, Jørgensen HE, Badger J, Hansen K, Ott S,
Rethore P, Larsen SE, and Jensen LE. The making of a 2nd generation wind farm model,
Wake Meeting 16th September 2008 at Dong Energy, Copenhagen.
[19] Larwood S. Wind Turbine Wake Measurements in the Operating Region of a Tail Vane.
NREL/CP-500-29132 2001
[20] Frandsen S, Thomsen K. Change in fatigue and extreme loading when moving wind
farms offshore. Wind Engineering 1997; 21: 197-214.
[21] Barthelmie RJ, Folkerts L, Ormel FT, Sanderhoff P, Eecen PJ, Stobbe O. Offshore wind
turbine wakes measured by SODAR. Journal of Atmospheric and Oceanic Technology
2003; 20:466-477.
[22] Christiansen MB, Hasager CB. Wake effects of large offshore wind farms identified from
satellite SAR. Remote Sensing of Environment 2005; 98: 251-268.
[23] Barthelmie R J, Larsen GC, Frandsen ST, Folkerts L, Rados K, Pryor SC, Lange B, Schepers
G. Comparison of Wake Model Simulations with Offshore Wind Turbine Wake Profiles
Measured by Sodar. Journal of Atmospheric Oceanic Technology 2006; 23: 888–901.
[24] http://www.offshorewindenergy.org
[25] Mosetti G, Poloni C, Diviacco B. Optimization of Wind Turbine Positioning in Large
Windarrays by Means of a Genetic Algorithm. Journal of Wind Engineering and Industrial
Aerodynamics 1994; 51:105-16.
[26] Jensen NO. A note of wind generator interaction, Roskilde, Denmark: Riso National
Laboratory 1993.
114
[27] Grady S. Placement of Wind Turbines Using Genetic Algorithms. Renewable Energy 2005;
30:259-70.
[28] Marmidis G, Lazarou S, Pyrgioti E. Optimal placement of wind turbines in a wind park
using Monte Carlo simulation. Renewable Energy 2008; 33:1455-1460.
[29] Mittal A. Optimization of the Layout of Large Wind Arrays Using a Genetic Algorithm.
Thesis. Case Western Reserve University, Cleveland, OH, 2010.
[30] Kusiak A, Song Z. Design of Wind Farm Layout for Maximum Wind Energy Capture.
Renewable Energy 2010; 35:685-694.
[31] Li S, Alexander J. Optimization of Wind Turbine Placement in Offshore Wind Arrays.
ASME International Design Engineering Technical Conferences 2011: Washington DC, US.
DETC2011-47935.
[32] Li S, Alexander J. Optimization of Wind Turbine Array Performance in Offshore Wind
Farms. ASME International Mechanical Engineering Congress & Exposition 2012:
Houston, TX, US. IMECE2012-87901.
[33] Wang F, Liu D, Zeng L. Study on computational grids in placement of wind turbines using
genetic algorithm. World Non-Grid-Connected Wind Power and Energy Conference 2009.
[34] Wan C, Wang J, Yang G, Zhang X. Optimal Siting of Wind Turbines Using Real-Coded
Genetic Algorithms. EWEC 2009.
[35] Huang H. Efficient hybrid distributed genetic algorithms for wind turbine positioning in
large wind arrays. IEEE International Symposium on Industrial Electronics 2009; iss. ISlE,:
2196-2201.
[36] Şişbot S, Turgut Ö, Tunç M, Çamdalı Ü. Optimal positioning of wind turbines on
Gökçeada using multi-objective genetic algorithm. Wind Energy 2010; 13: 297-306.
[37] Chen L, MacDonald E. A New Model for Wind Farm Layout Optimization with Landowner
Decisions. ASME International Design Engineering Technical Conferences 2011.
DETC2011-47772.
[38] Betz A. Introduction to the Theory of Flow Machines. (D. G. Randall, Trans.) Oxford:
Pergamon Press 1966.
[39] Gijs AM, Van Kuik. The Lanchester–Betz–Joukowsky Limit. Wind Energy 2007; 10:289–
115
291.
[40] Katic I, Hojstrup J, Jensen NO. A simple model for cluster efficiency. Proceedings of the
European Wind Energy Conference and Exhibition 1986; 407-410.
[41] Mortensen NG, Heathfield DN, Myllerup L, Landberg L, Rathmann O. Wind Atlas Analysis
and Application Program: WAsP 8 Help Facility. Risø National Laboratory, Roskilde,
Denmark.2005; 335 topics. ISBN 87-550-3457-8.
[42] Larsen GC, Madsen HA, Niels N. Mean wake deficit in the near field. European Wind
Energy Conference and exhibition 2003, Madrid, Spain.
[43] Schlichting H. Boundary layer Theory. McGraw-Hill Book Company 1968, Sixth edition.
[44] Ishihara T, Yamaguchi A, Fujino Y. Development of a New Wake Model Based on a Wind
Tunnel Experiment. Global Wind Power, 2004.
[45] Werle MJ. A New Analytical Model for Wind Turbine Wakes. FloDesign Inc., Wilbraham,
MA, 2008.
[46] Lissaman PBS. Energy effectiveness of arbitrary arrays of wind turbines. American
Institute of Aeronautics and Astronautics, 17th Aerospace Sciences Meeting, New
Orleans, 1979.
[47] Templin RJ. An estimation of the interaction of windmills in widespread arrays. National
Aeronautical Establishment. Lab. Report LTR-LA-171, Ottawa, 1974.
[48] Newman BG. The spacing of wind turbines in large arrays. Energy Conversion, 1977; 16:
169-171.
[49] Bossanyi EA, Whittle GE, Dunn PD, Lipman NH, Musgrove PJ, Maclean C. The efficiency
of wind turbine clusters. Proc. International Symposium on Wind Energy Systems,
Lyngby, 1980; 3: 401-416.
[50] Frandsen S. On the wind speed reduction in the center of large clusters of wind turbines.
Journal of Wind Engineering and Industrial Aerodynamics, 1992; 39: 251-265.
[51] Emeis S, Frandsen S, Reduction of horizontal wind speed in a boundary layer with
obstacles, Boundary- Layer Meteorology, 1993; 64: 297-305.
[52] Crespo A, Hernandez J, Frandsen S. Survey of Modelling Methods for Wind Turbine
Wakes and Wind Farms. Wind Energy; 1999: 1-24.
116
[53] Sforza PM, Stasi W, Smorto M, Sheerin P. Wind turbine generator wakes, American
Institute of Aeronautics and Astronautics, 17th Aerospace Sciences Meeting, New
Orleans, 1979.
[54] Taylor PA. On wake decay and row spacing for WECS farms. 3rd International
Symposium on Wind Energy Systems, Lyngby, 1980; 451-468.
[55] Liu MK, Yocke MA, Myers TC. Mathematical model for the analysis of wind-turbine
wakes. Journal of Energy 1983; 7: 73-78.
[56] Crasto G, Gravdahl AR. CFD wake modeling using a porous disc. European Wind Energy
Conference and Exhibition 2008, Brussels, Belgium.
[57] Ainslie J. Calculating the flowfield in the wake of wind turbines. Journal of Wind
Engineering and Industrial Aerodynamics 1988; 27: 213-224.
[58] Magnusson M. Near wake behavior of wind turbines. Journal of Wind Engineering and
Industrial Aerodynamics 1999; 80: 147-167.
[59] Rathmann O, Barthelmie RJ, Frandsen ST. Wind turbine wake model for wind farm
power production, European Wind Energy Conference, Athens, 2006.
[60] Barthelmie RJ, Pryor S, Frandsen ST, Hansen KS, Schepers JG, Rados K, Schlez W,
Neubert A, Jensen LE, Neckelmann S. Quantifying the Impact of Wind Turbine Wakes on
Power Output at Offshore Wind Farms. Journal of Atmospheric and Oceanic Technology
2010. 27: 1302-1317.
[61] "WindFarmer Website", Garrad Hassan & Partners Ltd, Accessed 2004, URL:
www.garradhassan.com/windfarmer/windfarmer.htm.
[62] Schlez W, Neubert A. New developments in large wind farm modelling. Proc. European
Wind Energy Conference, Marseilles, France, European Wind Energy Association, 2009.
[63] Schepers JG. ENDOW: Validation and improvement of ECN’s wake model. ECN-C-03-034
113 pp, 2003.
[64] Magnusson M, Rados KG, Voutsinas SG. A study of the flow down stream of a wind
turbine using measurements and simulations. Wind Eng. 1996; 20: 389–403.
[65] Politis ES, Rados K, Prospathopoulos JM, Chaviaropoulos PK, Zervos A. CFD modeling
issues of wind turbine wakes under stable atmospheric conditions. Proc. European Wind
117
Energy Conference 2009, Marseille, France, European Wind Energy Association, PO.163.
[66] Frandsen ST. Turbulence and turbulence generated structural loading in wind turbine
clusters, Risø-R-1188(EN), 2007.
[67] Dekker JWM, Pierik JTG, editors. European Wind turbine standards II.
[68] Anderson M. Simplified Solution to the Eddy- Viscosity Wake Model. Document
prepared by Renewable Energy Systems Ltd (“RES”), 2009.
[69] Crespo A, Hernandez J, Fraga E, Andreu C. Experimental validation of the UPM computer
code to calculate wind turbine wakes and comparison with other models. Journal of
Wind Engineering and Industrial Aerodynamic 1988; 27: 77–88.
[70] Schepers G, Barthelmie R, Rados K, Lange B, Schlez W. Large off-shore windfarms:
Linking wake models with atmospheric boundary layer models. Wind Engineering 2001,
25:307–316.
[71] Nielsen M, Jørgensen H, Frandsen S. Wind and wake models for IEC 61400-1 site
assessment. Wind Energy, EWEC 2009.
[72] Sempreviva AM, Larsen SE, Mortensen NG, Troen I. Response of neutral boundary layers
to changes of roughness. Boundary-Layer Meteorology 1990; 50:205-225.
DOI:10.1007/BF00120525.
[73] Kristensen L, Rathmann O, Hansen SO. Extreme winds in Denmark. Journal of Wind
Engineering and Industrial Aerodynamics, 2000. 87; 147-166.
[74] Larwood S. Wind Turbine Wake Measurements in the Operating Region of a Tail Vane.
NREL/CP-500-29132 2001.
[75] http://wis.usace.army.mil/wis.shtml
[76] Wind Statistics and the Weibull Distribution. Wind-power-program.com. Retrieved 11
January 2013.
[77] Eltra PSO-2000 Proj. NR. EG-05 3248. Wind Resources at Horns Rev, Program for
measuring wind, wave and current at Horns Rev, 2002
[78] Jensen L, Mørch C, Sørensen P, Svendsen K. Wake measurements from the Horns Rev
wind array. Elsam Engineering A/S Kraftvaerksvej 53, 7000 Fredericia.
118
[79] Davis L, Steenstrup M. Genetic search and simulated annealing. Lawrence Davis and
Morgan Kauffman, Los Altos, 1987.
[80] Goldberg, David E. Genetic Algorithms: in Search, Optimization & Machine Learning,
Addison-Wesley Publishing Company, Inc. New York, 1989.
[81] Nino R, Eecen P. ECN-DOWEC: Turbulence and wind shear: A literature study and
measurements. DOWEC- FIWI- PE-01-044/00, 2002.
[82] Barthelmie RJ, Frandsen ST, Rathmann O, Hansen KS, Politis E, Prospathopoulos J,
Schepers JG, Rados K, Cabezon D, Schlez W, Neubert A, Heath M. Flow and wakes in
large wind farms: Final report for UpWind WP8. Roskilde: Danmarks Tekniske
Universitet, Risø Nationallaboratoriet for Bæredygtig Energi. (Denmark.
Forskningscenter Risoe, 2011. Risoe-R; No. 1765(EN)).
[83] Ozturk UA, Norman BA. Heuristic methods for wind energy conversion system
positioning. Electric Power Systems Research; 70: 179-185, 2004.
[84] Christopher NE, James F, Mcgowan JG. Offshore Wind Farm Layout Optimization
(OWFLO) Project: Preliminary Results, Renewable Energy, p. 1-9.
[85] Bryony L, Jonathan C. An extended pattern search approach to wind farm layout
optimization. ASME International Design Engineering Technical Conferences &
Computers and Information in Engineering Conference IDETC/CIE 2010; p. 1-10.
[86] Christiansen MB, Hasager CB. Wake effects of large offshore wind farms identified from
satellite SAR. Remote Sensing of Environment 2005; 98: 251-268.
[87] Renkema DJ. Validation of Wind Turbine Wake Models Using Wind Farm Data and Wind
Tunnel Measurements. Master of Science Thesis. Delft University of Technology, 2007.
[88] Duckworth A, Barthelmie RJ. Investigation and Validation of Wind Turbine Wake Models.
Wind Engineering 2008; 32: 459-475.
[89] Kristensen L, Rathmann O, Hansen SO. Extreme winds in Denmark. Journal of Wind
Engineering and Industrial Aerodynamics 2000; 87:147-166.
[90] Larsen A, Larose GL, Livesey FM. Wind engineering into the 21st century. (pp. 243-250).
Rotterdam: Balkema Publishers, A.A. / Taylor & Francis The Netherlands, 1999.
[91] Frandsen ST, Barthelmie RJ, Rathmann O, Ejsing Jørgensen H, Badger J, Hansen KS, Ott S,
119
Rethore, P-E M, Larsen SE, Jensen LE. Summary Report: The Shadow effect of large wind
farms: measurements, data analysis and modelling: Risø-R-1615 (EN). (1 ed.) Risø,
Roskilde, DK: Risø National Laboratory, 2007. (Denmark. Forskningscenter Risoe. Risoe-R;
No. 1615(EN)).
[92] Milborrow DJ. The performance of arrays of wind turbines. Journal of Wind Engineering
and Industrial Aerodynamics 1980; 5:403-430.
[93] Hansen KS, Barthelmie RJ, Jensen LE, Sommer A. The impact of turbulence intensity and
atmospheric stability on power deficits due to wind turbine wakes at Horns Rev wind
farm. Wind Energy 2012; 15:183–196. doi: 10.1002/we.512.
[94] Pryor SC, Barthelmie RJ. Climate change impacts on wind energy: A review. Renewable
and Sustainable Energy Reviews 2010; 14:430-437.
[95] Rathmann O, Frandsen S, Nielsen M. Wake Decay Constant for the Infinite Wind Turbine
Array, EWEC 2010 Techn. Track “Wake”, paper ID 265.
120