WAKE INDUCED POWER DEFICIT ANALYSIS ON WIND TURBINES IN FORESTED MODERATELY COMPLEX TERRAIN USING SCADA DATA

(1)

Dissertation in partial fulfillment of the requirements for the degree of MASTER OF SCIENCE WITH A MAJOR IN WIND POWER

PROJECT MANAGEMENT

Uppsala University

Department of Earth Sciences, Campus Gotland

Esma Öztürk

(2)

Dissertation in partial fulfillment of the requirements for the degree of MASTER OF SCIENCE WITH A MAJOR IN WIND POWER

PROJECT MANAGEMENT

Uppsala University

Department of Earth Sciences, Campus Gotland

Approved by:

Supervisor, Dr. Karl Nilsson

Examiner, Prof. Jens Nørkær Sørensen

(3)

Abstract

Over the last few decades, wind power has shown a continuous and significant develop-ment in the energy market globally. Having reached a certain level in both technology and in dimensions, the role of optimizing wind turbines as well as wind farms has become an additional aspect to future development and research. Since turbine wakes can cause significant power deficits within a farm, research in this area has the potential for large improvements in wind farm design.

A wake is described as the downstream flow behind the rotor of an operating wind turbine. The two main characteristics of wakes are a velocity (momentum) deficit and an increased turbulence level. The velocity deficit behind the upwind turbine results in a power loss of the downstream turbines, whereas the higher turbulence causes additional loads on the downstream turbines’ structures resulting in fatigue problems.

The study of wakes is a complex topic, they are influenced by an interconnection of a number of parameters like ambient wind speed and turbulence, atmospheric stability conditions (stable, unstable, and neutral), the turbines’ operational characteristics, and the terrain properties.

In order to assess the power deficits affected by wake interaction between turbines, an analysis can be realized by processing SCADA data of turbines in a wind farm. The collected data is treated by a comprehensive filtration process, excluding events of icing, curtailment, faults, etc. and by grouping into different atmospheric conditions, wind speed intervals and wind speed sectors. Finally, power deficit values, as a function of wind direction, are calculated and quantified, and thereafter analyzed to assess the wake behavior at different conditions for different cases.

In this thesis, the wake-induced power deficit has been investigated in a specific study case for three pairs of two neighboring turbines in a forested moderately complex terrain using SCADA data. The production losses amounted between the range of 32 % to 67 % for the specific site with turbine spacing around 4 D. The obtained results were partially unsatisfactory, caused by the reasons of inaccurate wind direction values due to yaw misalignment issues and challenging separation into different stability conditions. Moreover, the power deficits showed a clear reduction of losses with increasing wind speed. A conclusion regarding the differences between stable and near neutral conditions could not be determined from the data.

(4)

Acknowledgements

First of all, I would like to express my deepest thanks to my Supervisor Dr. Karl Nilsson from Uppsala University who supported and advised me the entire time of my thesis, I appreciate that a lot. Also, special thanks to Prof. Kurt Hansen, his precious advice, and guidance. I would like to thank the whole Wind Department at Uppsala University, especially Prof. Stefan Ivanell, Dr. Hugo Olivares Espinosa, and Prof. Heracles Polatidis. Furthermore, I wish to express my sincere

thanks to Johannes Lindvall for his inspiration and support.

I am grateful to all of them for sharing expertise, and sincere and valuable guidance, and encouragement extended to me.

I would also like to use this opportunity to express my deepest gratitude and special thanks to my classmates in Visby. The year together with all of them was priceless! I wish to thank particularly Andis and Philipp; my project group Graeme, Jason, Jim, and Marc; and Frauke,

Abdul, Dimitrios, and Geoffrey for their support and motivation.

Therefore, I consider myself as a very lucky individual as I was provided with an opportunity to meet so many wonderful people in Sweden.

(5)

List of Tables

Table 5.1: Number of Data Points before and after Filtration . . . 28 Table 5.2: Power Deficit Results . . . 29

(8)

List of Figures

Figure 2.1: Power deficit distribution as a function of wind direction . . . 9

Figure 4.1: Flowchart describing the Methodology . . . 15

Figure 4.2: Sketch of farm layout . . . 22

Figure 5.1: Wind Turbine Power Signals T1 - Raw Data . . . 25

Figure 5.2: Wind Turbine Power Signals T1 - Remaining Data Points after Filtration 26 Figure 5.3: Wind Turbine Power Signals T1 . . . 26

Figure 5.4: Power Deficit Values Case 1 Near Neutral Conditions, Interval 1 -Observations . . . 30

Figure 5.5: Power Deficit Values Case 1 Near Neutral Conditions, Interval 1 -Mean Values and Curve Fit Function . . . 31

Figure 5.6: Power Deficit Values Case 2 Stable Conditions, Interval 2 - Observations 31 Figure 5.7: Power Deficit Values Case 2 Stable Conditions, Interval 2 - Mean Values and Curve Fit Function . . . 32

Figure 5.8: Power Deficit Values Case 3 Near Neutral Conditions, Interval 2 -Observations . . . 32

Figure 5.9: Power Deficit Values Case 3 Near Neutral Conditions, Interval 2 -Mean Values and Curve Fit Function . . . 33

(9)

List of Acronyms

CFD Computational Fluid Dynamics

D Rotor Diameters

IEC International Electrotechnical Commission LES Large Eddy Simulation

RANS Reynolds-Averaged Navier-Stokes RIX Ruggedness Index

SCADA Supervisory Control and Data Acquisition TI Turbulence Intensity

WRF Weather Research and Forecasting WTG Wind Turbine Generator

(10)

1 Introduction

1.1 Background

With a continuous development over the last few decades, wind power has achieved a significant position in the energy market competing with other energy sources. The great and fast improvement in design and efficiency of Wind Turbine Generators (WTGs) has lead to a new area of different research topics in the industry, where optimization has become an important aspect to future development [1]. Focusing on optimization, power losses lowering the energy yield and therewith the revenues have been tried to be addressed and quantified to find solutions for minimizing them [2].

Since power losses can have different origins as transmission inefficiencies, aero-dynamic under-performance, icing issues or several other reasons, losses caused by wakes have been paid special attention as wakes reflect a crucial physical topic which is complex and not fully understood so far [3]. In fact, turbines operating in the wake of other WTGs in a wind farm show a significant loss in energy production [4]. Barthelmie et al.[5] quantified these power generation losses by the order of 10 % to 20 % for large offshore farms.

Talking about wakes, two main issues affect wind turbine operation: the velocity (momentum) deficit of the upwind turbine resulting in a power loss of the downstream turbines and secondly an increased turbulence level causing additional unsteady dy-namic loads on the downstream turbines’ structures [6].

Planning a wind farm, wind resources are vitally important as are the WTGs’ technol-ogy and the suitability of the site. The economy focuses on the projects’ feasibility thus it does not allow high inaccuracies [7]. To reduce those uncertainties from the start and gain more knowledge in the physics and behaviour of the wind resources which affect the whole process of wind power development, power losses in wind farms caused by wind turbine wakes depict a significant topic in wind farm operation currently and notably more research in this field is required [5].

Wakes do not only affect the wind farm performance and energy losses, they cause higher loads on structures resulting in fatigue problems as well. Wind farm planning has to find a compromise between economics with respect to cost-efficient and feasible projects and farm layout decision considering limitations in space since the available pieces of land are not infinite. As a result, it is important to understand wakes and wake losses for accurate planning and estimations [2].

(11)

Not only the development of technology has made great process, but turbines and wind farms have also grown in size. Equally the market expanded from simple and flat sites to complex, forested and offshore areas [5]. As a result, wind farm modeling and planning faces new challenges, as previous methods deliver inaccurate under-predicted results regarding wake losses leading to an over-estimation of power production. This issue has an impact on both the economics as well as the grid integration of wind farms [5, 8].

Multiple types of models have been developed for the same approach, simulating wind turbine wakes and characterizing the downstream flow [6, 9]. Computational Fluid Dynamics (CFD) based on Large Eddy Simulation (LES) or Reynolds-Averaged Navier-Stokes (RANS) is counted among the variety of those computational models. However, most of the models’ results show high inaccuracies and variances regarding wake interaction and wake superposition [10], concluded in studies performing analyses with turbine Supervisory Control and Data Acquisition (SCADA) data [9].

Since the modeling of wakes is challenging, but of significant value for project planning and design, operation and performance evaluation [11, 12, 13, 14], research associated with minimizing energy costs by minimizing losses deals mainly with wind farm wakes [15].

Furthermore, energy losses are inferred from a complex interaction of different factors. Apart from wake properties, other physical phenomena like turbulence intensity and atmospheric stability affect the power production of WTGs. Wind direction and speed, turbine spacing, and operational characteristics have an influence on the wake propagation, as well [4]. Due to those wake losses amounting to relatively high values and the dependency on several parameters demands a comprehensive understanding of the wake flow and the interaction of complex flow structures [5, 16]. Additionally, the development of advanced models is needed as most current simulations for calculating the estimated power production are neither considering turbine wakes individually, nor their interaction with their surrounding [7].

Nowadays, the number of projects in rather complex than simple terrain is increas-ing [17]. Sites which are complex, forested and/or in cold climate show an additional challenge in the planning and realization process [3].

In order to understand the physical processes of wakes and wake interaction with ambient conditions for the design of a wind turbine and the layout of a wind farm, as well as for evolving comprehensive models, there is a need for investigation of this topic of wakes, for both, minimizing losses and loads [2, 18].

(12)

1.2 Objective and Research Questions

As already stated in the previous section, research in wakes and wake deficits is needed to identify physical properties as well as wake behavior in wind farms.

The main purpose of this thesis is to investigate the wake interaction between turbines in a forested moderately complex terrain by processing wind turbines’ SCADA data. Hence, this study aims primarily to assess the power deficit for a specific study case and secondarily comparing the results with literature. Since a standard for processing SCADA data does not exist for a power deficit analysis, this report focuses on the following research questions:

• How can the wake interaction of turbines in forested moderately complex terrain for a specific case be characterized?

• Do wakes dissipate due to terrain roughness and complexity and ambient turbulence intensity such that a power loss is not even perceivable?

• If wake losses are perceivable, how much is the power deficit caused by those wake losses? How much is it compared to other studies?

To summarize, the main objective of this thesis is to analyze the wake interaction in a forested moderately complex terrain and quantify the power losses caused by wake effects using SCADA data.

1.3 Outline

Outlining the organization of this report, the first three chapters of this thesis provide an overview and background information about wakes, its physical principles and current research in this field. The physics of wakes and wake effects, atmospheric stability conditions, turbulence, and the power deficit are described in chapter 2. The recent research and literature about wake deficits are reviewed in chapter 3. Completing the introductory part of this thesis, the applied methodology is discussed in chapter 4 including information about the processed data and essential information about the investigated turbines in this study. Chapter 5 follows with a presentation of the results and their discussion. Finally, chapter 6 focuses on concluding remarks and an outline for future work.

(13)

2 Theoretical Background

2.1 Introduction

This chapter aims to introduce the theoretical background for a better understanding of this report. Explaining the definition of terrain complexity, this section deals with the essential physical fundamentals of atmospheric stability conditions, wakes, and turbulence. Furthermore, information about the power deficit model and the type of data used in this study is provided.

2.2 Terrain Complexity

According to the International Electrotechnical Commission (IEC) Standards [19], a site is classified as complex if its topography shows substantial variations in features and obstacles inducing flow distortion. For a more precise determination of terrain complex-ity, an optimum or standard definition for a quantified value does not exist, anyhow there are few approaches modeling and indexing the area and surface properties [20]. Some modeling software tools use the Ruggedness Index (RIX) as a measure of terrain heterogeneity based on differences in site elevation [21, 22].

Terrain complexity shows relevance to wind energy, since wind varies not only temporally but also geographically, whereas the topography of a site has a significant impact on the wind climate. The ratio of land and sea, mountains and plains and the type of vegetation effects the wind characteristics due to variations in absorption and reflection of solar radiation, influencing the surface temperature and the humidity. Hills and mountainous regions are affected by variations in wind speed, as the wind flow is usually accelerated over and around hills and decelerated over sheltered valleys [7].

2.3 Atmospheric Stability Conditions

The atmospheric boundary layer represents the closest part of the atmosphere to the earth’s surface. Since physical values such as wind velocity, air temperature, and hu-midity vary in space and time, the characteristics of the atmospheric boundary layer are categorized in three different conditions determined by the vertical temperature distribution and the convective mixing of air. Thus, atmospheric stability conditions are

(14)

described as stable, neutral or unstable [11].

Atmospheric conditions are characterized as unstable, typically at daytime, when the heating of the sun causes warm air masses close to the ground to rise [17]. While rising, the air cools adiabatically as the surrounding air is not sufficient to cool the masses and achieve a thermal balance. Hence proceeding to rise, the air causes large convection and large-scale turbulent eddies resulting in a lot of vertical mixing [7].

Conditions are referred as stable, typically at night time or over waters, when the rising air masses achieve cooler conditions than the surrounding air and vertical mixing is suppressed [7].

The atmospheric stability is described as neutral when the thermal equilibrium of rising air to its surroundings is preserved. This occurs with strong winds causing large mixing of the boundary layer. In this case, the turbulence intensity is determined only by the surface roughness properties [7].

The classification of the atmospheric stability conditions for the use in research assign-ments is not that simple because ambient physical and thermodynamic knowledge is demanded. Especially determining the atmospheric conditions by needed measured val-ues as surface temperature and vertical temperature fluctuations, absolute atmospheric pressure, and specific humidity pose a challenge to obtain trustful results. A number of different calculation methods for the atmospheric classification used within simulations exist, like the Bulk - Richardson approach, the definition of the Monin - Obukhov length, or the internal Froude number, to name only a few [12, 23].

2.4 Turbulence

Turbulence describes the mixing of air originated by random fluctuations in wind speed within a relatively short timescale, based on a complex interconnection between surface topography and thermal effects in the atmosphere [24]. Those changes in wind speed propagate in all three directions: longitudinal (parallel to wind direction), lateral (perpendicular to the wind direction), and vertical [11].

The wind impinging on a wind turbine is turbulent. Likewise, due to the rotor plane extracting energy from the wind by changing the flow, additional turbulence is created. While the wind speed decreases due to the turbine, the turbulence increases behind the rotor [25]. Furthermore, when the rotor of a Wind Turbine Generator (WTG) is turning, torque occurs on the rotor axis and thrust on the rotor plane. Since turbulence leads to fluctuations in the aerodynamic forces, variations in mechanical loading cause additional fatigue loads. Designing a turbine, fatigue loading is one of the main factors influencing the lifetime and constructional design. For this reason, it is important to assess the physics of the wind and its characteristics [25].

(15)

In order to measure turbulence, the Turbulence Intensity (TI) is defined as the ratio of the standard deviation of the wind speed σuto its mean value u as shown in equation

2.1 for a 10minute time period [7, 11].

TI= σu

u (2.1)

Turbulence intensity changes with different atmospheric stability conditions [2], and regarding different wind speed levels, generally a higher TI is observed with lower wind speeds onshore [7]. The turbulence created by the wake of a turbine is traveling down-stream and dissipates until it obtains the ambient turbulence level of the surrounding [25].

2.5 Wakes

To extract the kinetic energy of the wind and generate electrical energy, a turbine is reducing the speed of the wind by decelerating the mass of air passing through the rotor disc [7]. Hereby the air flow is affected by the properties of the rotor, its size and the aerodynamic characteristics of the blades (mainly thrust coefficient) as well as its speed. The changed flow behind the rotor of an annular area is called the wake and its typical features are a reduced wind speed and increased turbulence resulting in energy losses [11].

The wake can be divided into sections of the near wake, the intermediate wake, and the far wake (beyond a distance of approximately 5 Rotor Diameters (D) from the rotor disk [24]) [25]. The region within the wake where the minimum wind speed is reached appears between 1 D to 2 D behind the rotor disk. In the far wake, the velocity profile can be described with a Gaussian distribution [24]. The shape and characteristics of the wake depend on turbine characteristics and ambient turbulence intensity governed by the terrain properties [11]. Wake recovery can go out to a distance of 10 D to 20 D depending on the atmospheric stability conditions [25]. Additionally, the position of a wake is not fixed; it is meandering, meaning that it moves both horizontally and vertically causing a velocity deficit and additional turbulence over a large volume of air [25]. Wakes are very complex and not yet fully understood.

2.6 Turbulence and Wake Dependency

Although both wakes and turbulence are different topics, they are interconnected. Wake caused turbulence is dependent on the rotor and blade characteristics (rotor diameter and blade chord), the rotor speed, the incoming wind velocity, topographical features and surface roughness, and thermal conditions [7, 25]. The ambient turbulence is influenced

(16)

by surface roughness and thermal conditions. Wake caused turbulence recovery is herewith governed by the ambient turbulence level [25].

Wakes by themselves are determined by a long list of factors, the WTGs operational conditions (tip speed ratio, blade pitch angle, yaw angle, power and thrust coefficients), rotor and blade geometry and properties, site topography and surface roughness, and wind conditions, whereby the thrust coefficient may has the highest impact [24, 26].

In terms of wind conditions, the TI has a significant influence on wake dissipation being faster with higher TIs occurring typically with lower wind speeds [7, 26]. The size and the recovery of a wake are furthermore characterized by atmospheric stability conditions [5, 7].

2.7 Turbulence and Wakes in Complex Terrain

In complex terrain, it is more difficult to quantify and describe the TI, since it is governed by the topography and surface roughness which lead to an increase of the ambient turbulence level. However, at the same time, the flow distortion results in a decrease of the TI [7].

Similarly for wakes, flow conditions over complex terrain differ from conditions over flat terrain caused mainly by local distortion effects. Those effects which are governed by the site specifications and thermal effects, wind conditions and flow characteristics are more challenging to determine [27]. Additionally, in forested terrain, the TI is relatively higher than the TI in flat terrain due to the surface roughness [17]. Therefore, a wake caused power deficit analysis on wind turbines in forested complex terrain is not as easy as an analysis in simpler areas [4].

2.8 Turbulence and Wakes in Wind Farms

Velocity deficits and additional turbulence caused by wakes have a significant effect on turbines in wind farms, resulting in power losses and higher structural loads [5]. Partic-ularly WTGs which are located in regions with superposition of wakes and turbulence caused by several turbines, suffer intensely [25]. Wind farm layout and turbine spacing display an important factor driving a power deficit caused by wake effects [7].

2.9 Power Deficit

Wakes and turbulence are important because the generated power of a WTG is sen-sitively dependent on the wind speed to the cube (eq. 2.2), meaning small changes in wind speed can cause huge deficits in power. For this reason, wind turbines try to achieve an optimum operational mode and are sensitive to any small amount of loss [24].

(17)

P= 1

2ρcpAu3 (2.2)

The equation above (2.2) describes the power output P [kW] of a wind turbine where ρ[kg/m3] represents the density of air, cp[-] the power coefficient, A [m2] defines the

rotor swept area, and u [m/s] characterizes the wind speed; with a maximum achievable value of 16

27 for the power coefficient (Betz - Limit). The equation clearly shows that for a small variation in wind speed results in a high variation in power output [7].

Power losses are dominated by the wind speed and a lot of parameters are intercon-nected to this value, the same applies to wake losses. The power deficit caused by wakes can be determined by calculating the power ratio of the two turbines, the one operating in the wake Pwake and the one facing the free stream Pf reeand afterward subtracting

the value from the number 1 [18]. The mathematical equation for the deficit η [-] is as follows:

η=1− Pwake Pf ree

; η ǫ [0;1] (2.3)

The power deficit can show values in the range between 0 and 1, whereas a value of 0 means that the downstream turbine is unaffected and shows no wake losses, a value of 1 would mean that the downstream turbine does not produce any energy at all due to wake shading. Depending on specific boundary conditions, the deficit can amount typically to a value between 10 % to 60 % based on literature [18].

It is important to note that different ways of determining the power deficit exist: the deficit could be described as a function of distance, time, surface roughness, atmospheric stability, ambient turbulence intensity, wind direction, et cetera. In this thesis, the power deficit as a function of wind directionis applied and talking about the deficit always means the deficit dependent on the wind direction.

(18)

Visualized (fig. 2.1), the power deficit distribution shows a Gaussian shape with its peak value at the wind direction when both turbines are aligned.

Figure 2.1: Power deficit distribution as a function of wind direction

Working with measured data from turbines, the calculated power deficit is naturally not a continuous function, it is more a set of observational points. To find the optimum function which fits those points the best, a curve-fit function can be applied. The function 2.4 describes the power deficit distribution as a function of the normalized wind direction θ[23].

f(θ) =a0+ (a1+a2∗θ+a3∗θ2) ∗exp(−a4∗θ2) (2.4)

The variables a0, a1, a2, a3, a4, and a5 can be determined by fitting the mean deficit

values from the available data.

2.10 SCADA Data

Modern turbines nowadays are all tracked by a SCADA system. A server enables operating data to be easily monitored through a remote interface, stored and archived for the later use. For each time stamp, the displayed and stored data mainly contains operational parameters like wind speed, wind direction, generated power, rotor speed, yaw angle and temperature as well as information about the state of operation and alarms of the turbine and various other parameters commonly representing 10 min average values [28].

(19)

3 Literature Review

3.1 Introduction

The number of publications in wake research is rapidly growing since wakes reflect an essential topic to both the fundamental physics of wind and to wind power project development. This chapter identifies the current research and studies in power losses caused by wake deficits in wind farms.

3.2 Current Research in Wakes

In the past, a number of research projects have been conducted to assess both, the physical characteristics of wakes and the energy losses in wind farms. There is a rapidly growing literature on wake research, however, a majority of the recent research concern-ing wake interaction in wind farms has mainly focused on offshore cases and less on onshore sites, where complex areas have been nearly neglected due to their difficulty [23]. Offshore wind farms such as Horns Rev, Middelgrunden, and Lillgrund are names immediately associated with wake studies.

One of the most popular projects that attracted a lot of attention by research teams is the offshore wind farm Lillgrund in the strait of Öresund between Denmark and Sweden [29]. Since the project was funded by the Swedish Energy Agency and was a pioneer project as the first Swedish large-scale offshore farm as well, information and data are publicly available to research teams [30]. The farm started operating in 2007, wind measurements for wake investigation have been used since the early stage. The wind farm consists of 48 turbines with a power of 2.3 MW each, and a rotor diameter of 93 m [30]. The remarkable fact is, as the permitting process for this offshore farm took 8 years [31], the initially planned turbine types where not produced anymore that a new and larger model had to be chosen. Since the agreed location of the single WTGs could not be changed due to the permission regulations, the farm ended up being built in the original layout with bigger turbines which resulted in a relatively dense configuration. Intending to reach maximum production instead of maximum efficiency, all initially planned turbines where erected instead of removing a few within the farm to reduce wake interaction. Therefore, wake losses have been expected to a significantly high amount [32] which provides an excellent basis for research. Additionally, the farm benefits from a met mast supporting the data from the farm. The assessment of the

(20)

power performance and wake effects showed a maximum peak loss of 80 % in power production for a distance of 3.3 D and a peak loss of 70 % for a distance of 4.3 D. Due to lack of information and data about atmospheric stability (temperature being only available at one level), this aspect could not be considered for the calculations [32].

Studies about the Middelgrunden Offshore Wind Farm using SCADA data have shown that wake caused power losses are the highest for lower wind speed ranges, explained with high thrust coefficients, and a decrease of those losses could be observed with increasing wind speeds. To assess quantities, the wind farm consists of 20 turbines with a power of 2 MW each, and a rotor diameter of 76 m. A power deficit analysis observed maximum peak losses of 80 % in power production for a distance of 2.4 D [2].

The Horns Rev farm also represents one of the farms where a lot of research campaigns have been carried out to assess offshore wake effects [5, 33]. The same findings as in the other studies are, that the wake deficit is higher for stable conditions than for neutral with the order of around 40 %, in this case for turbine distances of 7 D for a number of 80 turbines with a power of 2 MW each, and rotor diameter of 80 m [18]. In addition to the wake loss investigations, research to fatigue and lifetime assessment conducting aeroelastic load simulations has been realized [34]. Finally, the wake assessment still showed a too high degree of uncertainties with a remark on the yaw misalignment issues as well [35, 36]. From turbine to turbine investigations of wakes, farm to farm shading has also been started to be studied, investigating neighbored farms and their wake interactions to each other [37, 38]. Simultaneously, comparisons between different farms have been conducted as Horns Rev to Lillgrund [36, 39].

Besides studies with turbine and farm data, several studies modeling the wake in-teraction of wind farm have been conducted as well, for both wake research and more accurate model development [10]. Moreover, a number of campaigns initiated by the European Union can be found with the aim to characterize the flow and wakes over complex terrain and develop accurate modeling tools for an optimized wind farm devel-opment process [40]. Measuring and modeling wakes has been undertaken in programs like the EU-TOPFARM program [41] or the EERA-DTOC project [37, 42] to name a few. The programs commonly investigate existing models and software and assess their reliability by measurements to conclude future development [43, 44].

The offshore farm Lillgrund was used for LES simulations afterward comparing with the results from SCADA data [8] as well as for Weather Research and Forecasting (WRF) simulations [42]. Different models show different accuracies depending on the specific study which is carried out, WRF tends to overestimate the production [42]. CFD models using RANS equations also tend to overpredict wind speeds and resultingly underes-timate wake deficits [45]. Nevertheless, wake modelling is constantly developing and

(21)

improving, the LES studies conducted by Nilsson et al. [46] show a good agreement of simulations to the measurements of Lillgrund and several famous research groups who focused on wakes have published successful results as well.

In terms of wake physics, due to the number of studies with measurements as well as simulations, research obtained valuable knowledge about wake interaction in wind farms. The properties of wakes and wake behavior which was observed in the major-ity of the studies are, that wake and wake recovery depend on several ambient and local meteorological parameters and the interaction between wind speed, level of tur-bulence intensity and atmospheric stability conditions are complex [23]. Hence, the wake-induced power deficit is a function of both, meteorological and operational condi-tions as the wind speed, the superposition of ambient and turbine caused turbulence intensity, atmospheric stability conditions, turbine spacing and thrust coefficient [18]. Wake losses only perceivable below rated power, as the WTG operates with an optimum efficiency and a high thrust coefficient, reaching the level of rated power, a turbine regulates the power extraction of the wind by pitching its blades, therewith lowering its thrust coefficient which is indirectly coupled to the efficiency [7, 32]. Behind the rotor, a wake evolves downstream by increasing the turbulence intensity in length and time scale until it reaches its maximum and it decreases again until the wake is fully recovered [8]. Furthermore, wake recovery depends on turbulent mixing, determining the length of the wake [8]. The wake is strongly dependent on the ambient turbulence intensity, meaning with decreasing TI, typically at higher wind speeds, an increasing power deficit is observed. The same applies for the context of atmospheric stability conditions, since stable conditions usually show a lower turbulence intensity level, a higher wake deficit is observed when compared to neutral conditions. Supportively, higher wind conditions tend to head towards neutral conditions spoken for northern European waters [18, 23, 42, 47]. Focusing on offshore, the TI is also changing with a non-constant sea surface roughness varying with wind-induced waves, that with higher wind speeds high deficits are still perceivable, additionally influenced by relatively high thrust coefficients [2, 23]. Offshore, the lowest TIs are observed for wind speed ranges from 8 m/ sec to 12 m/ sec [18]. The turbine spacing within a farm is another significant factor. As a result, a decreased power deficit could be experienced for increased WTG spacing. A variation of the power deficit according to different wind directions was perceived [18]. The wake width for cases with typical turbine spacing was observed to lie in the range of 10° to 15° and is wider and lower at stable conditions due to less turbulent mixing [5, 18].

Focusing on onshore sites, the ambient turbulence levels tend to be higher when compared to offshore areas [2]. The complexity of the terrain coupled with the ambient turbulence level shows a great impact on how fast the turbine wake recovers [48]. With ambient turbulence levels being higher on onshore sites, the atmospheric mixing is

(22)

emphasized that the deficit variations in different atmospheric conditions may have a lower impact on the results [5]. Furthermore, it is assumed that wake effects in complex terrain can be governed by the distortion effects of the terrain topography and complex-ity, meaning that the local turbulence level may dominate the wake caused turbulence that power deficits are not even perceivable [27].

To conclude, a number of studies to assess wake characteristics exist already, however, there is still a need for further research [5], especially for onshore complex sites, as they show a relatively difficult topic and not enough studies have been conducted yet [1].

3.3 Summary

To summarize the findings of previous research, wakes are a complex topic with a number of parameters influencing them. The size and the way of recovery of a wake are determined by "the ambient wind speed and turbulence, the wake added turbulence, the turbine type, the terrain, the structure of the boundary layer relating to atmospheric stability, and the flow angle and variation with direction" [49]. The turbine type implies mainly on the thrust coefficient which is decreasing for increasing wind speeds [18] and therefore lowering the deficits as they are interconnected. A higher TI shows a lower deficit as well [9]. Moreover, the wake-induced power deficits incline to be higher in stable conditions than in neutral conditions offshore, and implementing the wake dissipation to be slower [18, 23]. Almost all conducted studies show relatively high wake losses, especially due to dense wind turbine spacing.

Nevertheless, the mentioned characteristics for wakes should be treated with aware-ness since offshore conditions differ to conditions onshore, so the results obtained in the studies may not be applicable for sites specifically in complex terrain as investigated in this thesis.

(23)

4 Methodology and Data

4.1 Introduction

The power deficit of turbines and the resulting production losses have been observed as crucial for the feasibility of a wind farm. In order to investigate the deficits affected by wake interaction between turbines, it is possible to realize an analysis by processing SCADA data of turbines in a wind farm. This chapter describes the applied methodology and the data used in this study [5, 18].

The methodology and filtering criteria have been derived with the help of the litera-ture and research of Barthelmie et al. [2][5], Hansen et al. [27], Hansson et al. [3], and the guideline of Sanz-Rodrigo et al. [50].

4.2 Methodology

This thesis has investigated a power deficit analysis for three pairs of neighboring tur-bines each by collecting data and wind farm information, performing a comprehensive data filtration and grouping process, and finally calculating discrete power deficit values for the analysis of the wake behavior at different atmospheric conditions at the specific site.

An overview of the study proceeding is illustrated in the flowchart in Figure 4.1. The methodological framework mainly consists of the three steps described as data collection, data processing, and power deficit calculations. Starting with data collection, this step is determined by the collection of WRF data, SCADA data, and wind farm information for the case study. The second step, data processing, is characterized by the data being quality-screened, filtered according to 11 criteria, grouped into 3 groups and lastly, the power values been normalized. Finally, the power deficit calculations deliver the wake deficit results for the discussion of this thesis. A more detailed description of the applied methodology and data treatment is specified in the following sub-chapters.

(24)

Figure 4.1: Flowchart describing the applied methodological framework considering data collection, data processing, and the power deficit calculations

(25)

4.2.1 Data Collection

The power deficit analysis in this thesis has been conducted for six turbines of a wind farm, but only two neighboring WTGs at a time, for three cases in total. The following sections describe the process for a case in general terms.

The upwind turbine has been chosen as the reference turbine called WTG A and the downwind turbine as the object turbine referred as WTG B. More detailed information about the farm and the site is described in section 4.4.

The collected data mainly consists of SCADA data, WRF data and wind farm in-formation. The wind farm information (section 4.4), is needed to do an assessment with the results of the deficit calculations. The WRF Model data is data provided by mesoscale numerical simulations by a weather prediction system [3], delivering atmo-spheric stability values used for the grouping of different ambient conditions before performing the calculations. The SCADA data forms the main part of the processed data set. It is provided by a wind farm experiencing wake interaction. The required and provided parameters tracked by the WTGs systems used in this study for the power deficit analysis are as follows:

• Time Stamp (Date) [mm/dd/yyyy hh:mm] • Electrical Power (P_ACT) [kW]

• Nacelle Yaw Position (POS_NAC) [°] • Nacelle Wind Speed (V_WIN) [m/s] • Ambient Temperature (T_AMB) [°C] Status messages:

• Operating State (OS) [scale 0-20]

• State and Fault (STATE_FAULT) [scale 1-16] • Turbine State (WTGState) [scale 0-5]

As typical, all parameters were registered as 10 min statistical mean values with the data of both WTGs having the same format and units.

4.2.2 Data Processing

To quantify the power losses caused by wake effects, it is important to consider data for the power deficit analysis which reflects time periods of normal operational conditions. Events like icing, curtailment, faults and general disturbance should be filtered out to preserve representative study results. Especially icing events as they cause additional power losses and it is not possible to separate them from the wake losses in the final

(26)

results [51].

The amount of data first needed to be processed to provide the basis for the deficit calculations. The qualification of data, or referred as quality screening as well, marked and excluded outliers and spikes, drop-outs and erroneous observations, followed by the synchronization of the data set for both turbines. Only data points which were available for both turbines for the same time stamp have been further processed. Statistics were made to display available data points. Subsequently, the data was filtered according to following criteria to ensure representative power deficit calculations:

1. Periods of wind turbine events (start or stop)

• Exclusion of all time stamps and data points with SCADA status messages indicating aforementioned events

2. Periods below cut-in wind speed/power including cut-in/cut-out hysteresis • Exclusion of all time stamps and data points with mean wind speed less than

4 m/s and electrical power production below mean production at 4 m/s wind speed

3. Periods above rated power

• Exclusion of all time stamps and data points with mean wind speed greater than 12 m/s electrical power production above mean production at 12 m/s wind speed

When a WTG reaches its rated power, its blades start pitching out to reduce the thrust coefficient, which is indirectly a measure of how much energy the turbine extracts from the wind. Since the WTG regulates the energy production additionally by operating with constant power as well, a deficit would be less pronounced. 4. Periods with icing

• Exclusion of all time stamps and data points with a temperature at hub height below 3°C [3] and values with a difference to the mean power production higher than 1 σ standard deviation

Wet snow can occur to a temperature up to3°C [52]; the probability density function of the wind speed within a time period of 10minutes is assumed to be normally distributed (Gaussian),68 % of the values are expected to be within the 1 σ range [25].

According to the IEC task 19, icing events can be filtered out for values with10 % variance from power curve and temperatures below0°C.

5. Periods of idling

(27)

• Exclusion of all time stamps and data points with SCADA status messages indicating aforementioned event

6. Periods of curtailment

Alternatively, the power values could have been checked for high deviation from the operational power curve and thereafter be excluded.

7. Periods of special events (updates, calibration, retrofitting)

• Exclusion of all time stamps and data points with SCADA status messages indicating mentioned events

8. Periods of grid disconnection and grid faults/failures (both turbines must be grid connected 100 % of the time)

9. Periods of alarms, fault conditions, service and maintenance actions, configuration changes, manually shut down times

• Exclusion of all time stamps and data points without normal operation mode in the SCADA status messages

Further filtration for specific study case: 1. Periods outside of flow deficit sector

• Exclusion of all time stamps and data points outside of flow deficit sector Which first had been set to±15° around the turbines’ alignment direction to cover the wake [5], but afterwards due to misalignment of the instrumentation a shift of the measured direction to the actual direction was visible, which became apparent during the calculations, therefore a new sector of±30° around the alignment direction was applied not to exclude relevant data points.

2. Periods of residual turbines influence (critical zones were checked on the map with farm layout)

• Automatically done with the criterion afore

Having finished the filtering process, the data were categorized into groups. These groups were build according to the following 3 criteria:

• Different atmospheric stability classes (stable, unstable, neutral with information from WRF simulations)

(28)

• Wind direction sectors of 60°

Afterwards before the deficit calculations, once again statistics were created about how much data is left after the filtration process, and if the amount of data is still repre-sentative for further assessment.

Lastly, as both investigated turbines have different rated power levels (see section 4.4), the power values have been normalized to a maximum value of 1 each to perform the final calculations, the power deficit ratios.

4.2.3 Power Deficit Analysis

The power deficit as a function of wind direction was calculated according to equation 2.3. Afterwards, mean values for each wind direction bin of 1° were determined using a 5° moving average window, that power losses due to wakes could be quantified with the help of a curve-fit function 2.4. It is important to note, that the wind direction needed to be normalized to perform the curve-fit. The 5° window was chosen for two reasons: due to a small amount of data to build a more representative statistical mean value, and the instrument uncertainty is expected to be less than 5° according to the IEC standards, so an accuracy of 1° would be unrealistic. Additionally, the yaw misalignment and offset (later explained in section 4.4) could be ascertained with the help of the power deficit curves as the peak was expected to appear at the wind direction where both turbines are fully aligned.

Results in form of plots of the power deficit curves are displayed in the following chapter (5), and a comparison to other studies appears in chapter 6.

4.3 Data

For the assessment of the power deficits caused by wakes, SCADA data was accessible from another research project of a specific wind farm for the time period of [01.09.15 00:00 to 23.6.16 00:00]. All WTGs had data available for the same time period, with occasionally missing data in between different time steps, by the time stamps representing 10 min average values each. The utilized WRF data was provided by a third-party who was involved in the mentioned research project. Furthermore, the wind farm information was available from the project developer.

An offset correction, adjusting the raw data by the offsets of different instruments, could not be performed due to lack of information about instrumentation.

As the reference wind speed, the wind measurement data by the anemometer of the upwind turbine WTG A was used, since it was not possible to derive it by the power performance of the turbines. The provided power curve from the manufacturer is for different conditions, despite an air density correction, the operational power curve

(29)

deviated a lot from the power curve given by the manufacturer. Subsequently, it was decided to use the SCADA data for the wind speed instead. In addition, as the power curve from the manufacturer deviated significantly compared to the measured data points, the operational power curve was calculated using the provided data and used for further proceedings.

Furthermore, the nacelle wind direction was not included in the given dataset. There-fore, the nacelle yaw position from the SCADA data from the upwind turbine WTG A was taken as reference for the wind direction.

The dataset giving information about the atmospheric stability classes is based on the static stability over the surface vertical layer, determining the potential temperature difference over vertical sections, with the diurnal development been taken into account [53].

4.4 Site Description

4.4.1 Introduction

For this thesis, a power deficit analysis for three pairs of two neighboring turbines each was conducted assessing SCADA data from a medium-sized wind farm located in Sweden. Due to confidentiality reasons, the case study will only be described in general terms; hence only necessary information of the site and the turbines will be provided for the comprehension of this report.

4.4.2 Wind Farm Information

The investigated wind farm consists of several conventionally available WTGs, all three-bladed, horizontal axis and pitch regulated, variable speed turbines. The farm has been in operation for only a couple of years. The six analyzed turbines are named as T1, T2, T3, T4, T5,and T6 located in an outlying position to the remaining turbines. Case 1 in the study assesses turbines T1 and T2, Case 2 investigates the WTGs T3 and T4, and finally, Case 3focuses on the WTGs T5 and T6.

The prevailing wind direction at this location is from South-West. The examined inflow angles for the cases when each pair of turbines is fully aligned and in the shade of the wake of the upwind turbine, and at the same time the cases not influencing each other, are as follows:

• Case 1: 204°±30° • Case 2: 290°±30° • Case 3: 187°±30°

(30)

The wind farm expands within an area of 3 km * 1 km by a relatively dense layout with a spacing between the turbines of around4 D (Case 1: 4 D, Case 2: 3.8 D, Case 3: 4.3 D).

The terrain itself could be described as complex as it represents moderately complex topography including forested area [20]. However, the delta RIX amounts only to a value less than 1 % [53]. According to the IEC Standards’ [19] definition anyhow, the terrain is classified as complex as it is revealing topographic variations by the highest elevation difference of the turbines within the farm amounting to 27.4 m and the lowest variation in height accounting for 0 m. To be more precise, if T1 is taken as reference with a height of 0 m, the height difference to T2 is−6 m, to T3−8.7 m, to T4−5.9 m, to T5−10.8 m, and to T6−27.4 m, by T1 been situated at the highest position above the other turbines. To summarize, the height differences of each case from the upwind to the downwind turbine are: Case 1: −6 m, Case 2: 2.8 m, Case 3:−16.6 m.

Additionally, the site mainly consists of forests with a few patches of agricultural land. The forest land is dominated by coniferous forest varying in age and thereby varying in height (8 m to 12 m reported by the database of Swedish Forest Agency [53]). Some small areas of the forest have been harvested as well [54].

To have a look into the technical details, turbines T1, T4, T5, and T6 have a rated power of 2.750 kW reached at 13 m/s wind speed and WTGs T2 and T3 a rated power of 2.470 kW reached at 12 m/s wind speed. The cut-in wind speed and the cut-out wind speed have the same values for all turbines 3 m/s and 25 m/s. Moreover, all WTGs have a rotor diameter of 103 m and a hub height of 98.5 m. It is worth noting that T2 and T3 are operating at certain times in noise reduction operation mode to lower acoustic emissions [55].

A summary of the wind farm information and a sketch of the farm layout is visualized in the following figure 4.2.

(31)

Figure 4.2: Simplified sketch of farm layout - top view

4.5 Uncertainties

Working with data can entail uncertainties which should be made aware of and con-sidered interpreting results obtained with those data. To begin with the amount of operational data, the available amount of SCADA data was limited, meaning the pro-vided data set only consisted of a time period of fewer than 10 months instead of at least one year. Hereby the summer months were missing, which omits the seasonal influence on the wind distribution. The power curves are not fully representable for this reason since they usually represent a one year period [3].

Additionally, the data was expected to be reliable as the majority of the study was performed by using the status information and the wind speed and direction data of the SCADA system. It was trusted on the accuracy of all the instrumentation and the nacelle transfer function which can create a bias in the wind speed data as well.

Another assumption is that by using the nacelle yaw position of the upwind turbine, the wind faces the rotor perpendicular and it cannot be assured that the downwind turbine is facing the same direction. It can not be asserted that for both investigated turbines the same meteorological and atmospheric conditions obtain.

(32)

Moreover, calculated power losses are considered solely as wake losses, whereas they can be partially originated by other sources as well. Due to wake losses usually very much greater than losses caused by other reasons, those other factors are neglected.

To address may the most challenging issue, the accurate determination of wind direction. The nacelle yaw position which has been used as reference wind direction will introduce some additional uncertainty. The nacelle yaw position is biased with both, a presumably constant instrumentation offset owing to miscalibration and a definitely non-constant yaw misalignment. The yaw misalignment can occur due to the turbines’ control system and inertia since the wind naturally shows short-time variations and the turbine tries to yaw towards the inflow direction [45]. This observation results in a yaw error, whose ratio of constant offset and non-constant misalignment is hard to determine [49]. Previous studies dealt with the same challenge detecting a systematic deviation [15, 49] observing values up to several tens of degrees [56] due to sensor inaccuracies measurement noise, poor resolutions, and the bias and offsets. Nevertheless, the wind direction variations within the 10 min average time period considered to be normally distributed, the yaw error should even out as it is averaged as well and expected not to be that high and close to zero [45]. For this reason, the data is relied on and in fact, the deficit calculations are based on power value for same time stamp, not the same yaw direction, the relative direction is only used for plotting the results.

(33)

5 Results and Discussion

5.1 Introduction

In order to investigate the wake-induced power deficit for a specific site which is described as forested moderately complex, SCADA data has been processed for three pairs of two neighboring turbines. The data processing has been performed as described in the methodology (ch. 4). After a comprehensive filtration and grouping process, the power deficit values could be determined for the different cases in different atmospheric stability conditions for three different wind speed intervals. Due to lack of data in unstable conditions and unsatisfactory statistical representativeness, only the results for stable and near neutral conditions will be presented and discussed in the following sections.

It should be noted that the complete result plots can be found in appendix A.

5.2 Raw Data

Before starting with the data processing, productional data from each turbine was separately visualized and analyzed. The measured power values for T1 are displayed in figure 5.1, the simultaneous plots for all the other turbines can be found in appendix A.

The raw data shows a high diffusion, several observational points when the turbine is on standstill and curtailment levels can be seen from the plot as well. Nevertheless, the shape of the power curve is perceivable.

(34)

Figure 5.1: Wind Turbine Power Signals T1 - Raw Data

5.3 Filtered Data

The power signals from the turbines SCADA system have been filtered to exclude observations with erroneous data, curtailment, icing and all other events indicating that the turbine is not working under normal conditions. Furthermore, time periods where both of the turbines in each case were not available at the same time were also filtered out. Again, for T1 the filtered data and a plot of the data points which were excluded are illustrated in the following plots 5.2 and 5.3. Plots for the other turbines can be found in appendix A.

(35)

Figure 5.2: Wind Turbine Power Signals T1 - Remaining Data Points after Filtration

Figure 5.3: Wind Turbine Power Signals T1 - Excluded Data Points Marked in Magenta, Remaining Data Points Marked in Black

The filtered power values still show a broad distribution which can be explained by the terrain properties, its complexity and the roughness of the forest. Additionally, the values are not grouped into stable and neutral conditions and it should be kept in mind

(36)

that the values represent not a whole year.

To assess statistical representativeness, the number of observations before and after the filtration and grouping process was inspected and listed in table 5.1. The number of data points per stability class, wind speed interval and wind direction seem relatively low, however, a sufficient number of data to perform the deficit calculations exists. It is noticeable that the number of observations for near neutral conditions clearly sur-mount the number of data points for stable conditions. According to Machefaux et al. [12], onshore, the conditions tend to be rather neutral than stable, which supports the distribution of data in this thesis, however, it is mentioned in this source that there are some discrepancies and near-neutral conditions are experienced only a fraction of the time.

Furthermore, the amount of data decreases with increasing wind speeds for all cases and although the prevailing wind direction is from south-west, and the cases focused on different wind direction sectors, the amount of data points for each case is comparable with Case 2 showing an unexpected higher amount than the other two cases despite being the only case not concentrating on the main wind direction.

The number of observations for each stability class and wind speed interval in the particular wind direction sectors varies from 48 to 1.913 which in any case represents less than 1 % of the whole time period.

(37)

Table 5.1: Number of Data Points before and after Filtration for all three Cases; grouped in near neutral and stable conditions, as well as for wind speed intervals and

wind direction sectors

Case 1: T1 & T2 Case 2: T3 & T4 Case 3: T5 & T6

Total number of

raw data points 42,124 42,061 42,371

near neutral stable near neutral stable near neutral stable 1. Whole time period

(01/09/2015 - 23/06/16) 296 days∼10 months

23,829 12,688 23,614 12,761 23,911 12,772

2. After filtration 12,896 6,686 17,087 8,373 14,796 7,883

Wind speed Intevarls I1 - I3

I1: 5 - 7 m/s 4,324 3,070 6,150 3,763 5,630 3,808

I2: 7 - 9 m/s 3,171 1,681 4,122 1,577 3,018 1,179

I3: 9 - 11 m/s 1.561 328 1,558 216 1,191 202

Intervals I1 -I3 to specific wind direction

sectors of each case

I1-S: 5 - 7 m/s 1,125 722 1,913 955 1,145 901

I2-S: 7 - 9 m/s 1,103 550 1,762 548 932 319

I3-S: 9 - 11 m/s 431 105 616 97 250 48

5.4 Power Deficit Analysis

The power deficit values were calculated following the equation 2.3 after the data had been filtered and grouped to different stability conditions, wind direction sectors and wind speed intervals. Hereafter a mean value was determined per each wind direction bin of 1° using a 5° moving average window. The mean values were afterwards used for the curve fit following the equation 2.4 with results for the unknown coefficients a0to a4

with a confidence bound of 95 % (MATLAB). By determining the peak values for each power deficit curve, the maximum losses could be achieved. The obtained results for all the three assessed cases are listed in table 5.2. The values marked in red indicate the conspicuous values discussed in the following sections.

(38)

Table 5.2: Power Deficit Results: Peak Values of Curve Fit for near neutral and stable conditions, grouped into wind speed intervals

Case 1: T1 & T2 Case 2: T3 & T4 Case 3: T5 & T6

Wind speed

Intervals I1 - I3 near neutral stable near neutral stable near neutral stable I1: 5 - 7 m/s 0.6431 0.6194 0.5741 0.6122 0.6487 0.6498 I2: 7 - 9 m/s 0.5596 0.6059 0.4876 0.5054 0.5611 0.6709 I3: 9 - 11 m/s 0.3159 0.3608 0.3202 x 0.4146 0.5519

Starting with Case 1, the deficit decreases with increasing wind speed for the near neutralcase from 64 % to 32 % and for the stable case from 62 % to 36 %. Comparing the two atmospheric conditions, the deficit values do only show a slight difference to each other. Looking at the first interval of 5 m/s to 7 m/s wind speed, the deficit values for the stable case seem lower than for the near neutral case, although the opposite would be expected. The deficit values for the higher wind speed ranges show higher losses for the stablecase.

Continuing with Case 2, as for the previous case, the deficit decreases with increasing wind speed. For the near neutral case peak losses from 57 % to 32 % and for the stable case deficits from 61 % to 50 % are observed, with the exception of the wind speed interval 3 not being able to be determined due to lack of data in this case. Generally, the power deficits are higher in stable conditions than in near neutral.

For the last case, Case 3, in near neutral conditions the deficit decreases with increasing wind speed and ranges from 65 % to 41 %. The same cannot be said for the stable case. Here the values do not show a trend, starting with 65 % for the first wind speed interval, the deficit increases for the second interval to a value of 67 %, decreasing again for the highest speed interval to 55 %. Furthermore, it is conspicuous that for the first wind speed interval of 5 m/s to 7 m/s the deficit values in stable and near neutral conditions are almost the same.

Furthermore, it was observed that for all three assessed cases, the wake centers were shifted and not appearing at the expected geographical angles. In Case 1, according to the geographic position, T1 and T2 are fully aligned when they face an angle of 204°. The deficit peak occurred at an angle of 190° (−14° shift). Case 2 was expected to have the deficit peak at 290°, actually, the peak appeared at 268° (−22° shift). For Case 3 the geographical alignment direction amounts 187°, the peak value appeared at 180° (−7° shift).

Trying to come to a conclusion, all three cases nearly show the same trend: the power deficit decreases with increasing wind speeds. This can be explained by both the TI and

(39)

simultaneously the thrust coefficient decreasing with higher wind speeds [13]. Focusing on the differences between the different atmospheric conditions, a trend for stable and near neutralbeing higher, lower or the same, cannot be observed.

The results not being that satisfactory, it is taken a deeper look into the observational points before the curve fits, as the results in table 5.2 are derived from the curve fit.

As an example, the deficit for Case 1, interval 1 in near neutral conditions looks as expected, it shows a high scatter but still forms a nice shape for the mean values and the curve fit (figure 5.4 and 5.5).

Figure 5.4: Power Deficit Values Case 1 Near Neutral Conditions, Interval 1 -Observations

(40)

Figure 5.5: Power Deficit Values Case 1 Near Neutral Conditions, Interval 1 - Mean Values and Curve Fit Function

If we take another case, for example in Case 2 for stable conditions the second wind speed interval, the data points show an unacceptable high scatter that a deficit curve shape is not even perceivable (figures 5.6 and 5.7). Calculating the mean values, which show two peaks, and determining the curve fit function obtains results, those results should be treated with care.

(41)

Figure 5.7: Power Deficit Values Case 2 Stable Conditions, Interval 2 - Mean Values and Curve Fit Function

Another interesting result shows the Case 3 for near neutral conditions in the second wind speed interval (figures 5.8 and 5.9).Here it looks like two deficit curves are overlap-ping with a shift in the plot with the data points only (figures 5.8), but that is not visible in the plot with the mean values for the curve fit (figures 5.9).

Figure 5.8: Power Deficit Values Case 3 Near Neutral Conditions, Interval 2 -Observations

(42)

Figure 5.9: Power Deficit Values Case 3 Near Neutral Conditions, Interval 2 - Mean Values and Curve Fit Function

Conclusively, the deficit calculations and plots for Case 1 look satisfactory, but the observational points for the deficit in Case 2 as well as in Case 3 do not look tolerable although the curve fit delivered results, a disagreement is observed. A statistical problem can be excluded since all three cases have a similar amount of data available. Therefore, it is questioned if those results are representative as they seem to be deceptive. This may show that the curve fit is not that sensitive. An extension of the moving window in the mean value calculations would smoothen out the distribution but still, falsify the results.

Reviewing literature, previous research already dealt with challenges in wake deficit calculations. Terrain complexity, scatter in the results and lack of data for specific wind direction sectors and wind speed ranges make it challenging to quantify wake losses [5, 18].

To address possible reasons for the suspicious results, the first thing coming in one’s mind is, if any yaw error or calibration happened during the assessed time period, or if the yaw misalignment and offset is that high that in some plots it looks like two deficit distributions are shifted and overlapping. Although a sector of 60° wind direction was assessed, the deficit focused only on a 30° angle which is typical width for a wake in the given turbine spacing [5]. So that should not have influenced the single wake observation. For Case 1 and Case 2, there cannot be an influence or wake superposition by the remaining turbines due to the geographical position. Case 3 might has been affected by the turbines of Case 1. There is no doubt about the calculated power deficit values being wrong as they were always calculated independently from the wind direction, the ratios were always built for the same time period. Only when visualizing and trying to find an optimum fit, the deficits are depending on the wind direction values. The case of any recalibration of the direction sensors that a fraction of the data points is shifted, information about such an event is not available. The yaw misalignment can also falsify the actual wind direction since it is not constant. If it shows a high variation that would

(43)

explain a high scattered plot as well. Previously, for all three investigated cases wake center shifts had been observed as relatively high. In fact, the yaw misalignment had been addressed as one of the biggest challenges in wake analysis [36].

Another factor is that the wind turbines in this case study are all situated in very closes distances to each other. Following the literature, they are located in the inter-mediate wake region since they show a distance equal to or less than 4 D to 5 D [6, 24], intermediate wake conditions and wake meandering might influenced the data set as well. A non-linear interaction in the near wake had been observed previously, considering tower shading [57].

It is not expected that both turbines face the same wind direction, but the upwind turbine is always taken as a reference that it is made sure that the downwind turbine is experiencing the wake. Hence the assumption that the reference turbine is facing the wind direction perpendicular, specifically knowing that there is a yaw misalignment, may cause a deception.

Another factor to be critical with is the separation of different atmospheric stability conditions. In the results there is no clear trend or connection between stable and near neutralconditions, the results look randomly distributed. The classification values could be biased, but anyhow they should not have influenced the final results significantly since for onshore sites as atmospheric mixing is emphasized due to terrain conditions, stability influences are observed to be reduced [5]. Only in offshore cases, studies reveal a clear tendency of power deficits being distinctly higher in rather stable than in neutral conditions [12] dominated by turbulence intensity levels. Hence the wake dependency on the ambient turbulences may be stronger than on stability conditions, as the power deficit distribution is often also defined as a function of turbulence intensity. Moreover, Barthelmie et al. [35] as well as Sanz-Rodrigo et al. [50], have observed that the way of treating and processing data impacts the results remarkably and reveals uncertainties. Thereby high standard deviations seem normal. The results of this thesis also show a large standard deviation, but this should apply to all cases then and does not explain why the cases differ from each other.

Additionally, it has been observed that different wind direction entails different conditions [23, 33], the three cases were assessed for different sectors. Consequently, the atmospheric conditions for each case could be biased, as they are not simulated for each direction of the wind farm, more for each time stamp. Also, the thesis focused mainly on ambient conditions for the turbines, those conditions can vary in small scale from one turbine to the other turbine as it is a complex interaction between several parameters and wind and turbulence conditions change from the upwind to the downwind turbine, meaning that if the first turbine experiences neutral conditions, the second one can