Professional Documents
Culture Documents
d
Praa Marechal Eduardo Gomes, 50, Vila das Accias, CEP 12.228-900, So Jos dos
ite
Campos, SP, Brasil.
adaiana1@yahoo.com.br
ed
Edson Luiz Zaparoli
py
Technological Institute of Aeronautics (ITA), Mechanical Engineering Division
Praa Marechal Eduardo Gomes, 50, Vila das Accias, CEP 12.228-900, So Jos dos
Co
Campos, SP, Brasil.
elzaparoli@gmail.com
Cludia R. Andrade ot
Technological Institute of Aeronautics (ITA), Aeronautical Engineering Division
tN
Praa Marechal Eduardo Gomes, 50, Vila das Accias, CEP 12.228-900, So Jos dos
Campos, SP, Brasil.
claudia@ita.br
rip
sc
ABSTRACT
nu
Microscale numerical modeling is currently the main tool used in wind industry to assess local wind
Ma
resources. This paper presents a systematic procedure to adjust CFD predicted wind profiles to
experimental measurements in order to minimize their differences. It can be applied when wind
ed
measurements are available. Data from 10 masts with several measurement heights from the well-known
pt
Bolund hill experiment provided the observed wind profiles. Simulated profiles were calculated with
WindSim CFD model for the aforementioned site. Speed-up correction factors were defined through the
ce
Least Squares Method to cross-correlate each mast as reference to all the others inside the Bolund hill
Ac
domain. After, the observed and the adjusted wind profiles at the same position were compared.
Moreover, RMSE errors were used as a metric to evaluate the estimations and the ability of each position
to be predicted and predictor. Results have shown that the quality of the adjustment process depends on
1
Corresponding author.
SOL-16-1465, SILVA 1
the flow characteristics at each position related to the incoming wind direction. Most affected positions,
i.e. when the airflow overcomes the Bolund hill escarpment, present the less accurate wind profile
estimations. The reference mast should be installed upstream of the potential wind turbines locations and
d
ite
Keywords: Renewable energy, wind engineering, wind profile, speed-up, CFD
ed
1. INTRODUCTION
py
Face to the current global scenario of fossil fuels scarcity for power generation,
Co
the use of renewable sources has significantly increased [1]. Wind energy is gradually
ot
becoming one of the most promising alternative sources of energy due to its cleanliness
tN
wind speed is a crucial issue in the wind power industry [2-4]. The selection of an
r
sc
inappropriate area can directly jeopardize the project profitability and lead to the failure
nu
of the investment. Hence, the motivation of the present study came from the need of
Currently, numerical modeling of atmospheric flow is the main tool adopted for
ed
local flow distribution studies and, in many cases, steady-state treatments are feasible
pt
for solving the time average flow and estimating turbulence statistics. Theoretically this
ce
is valid when the region of the terrain under investigation is small enough so that the
Ac
time required for an air particle to pass through the region is much smaller than the
SOL-16-1465, SILVA 2
steady-state flow in order to obtain 3-dimensional wind resource maps [6-11]. CFD
(Computational Fluid Dynamics) models, such as the one employed in this study, solve
d
ite
characteristics (velocity, pressure and temperature). These models generate a basic
ed
wind field considering local terrain perturbations and standard inlet boundary
py
conditions. Since they dont take into account the actual local speed values and velocity
Co
directions, the CFD results need to be accordingly adjusted. This adjustment is made by
taking measurements in one or more points inside the considered region, estimating the
ot
velocities to other points identified as potential wind turbine locations internal to the
tN
domain [7, 12]. The measurements are normally provided in frequency distributions of
ip
Within this context, this paper employs statistic approach based on the Least
nu
Squares Method to fit vertical wind profiles predicted by a CFD model in distinct
Ma
positions of the studied area, minimizing the difference between simulation and
create a relationship between the flow at the measurement point and its surrounding
pt
points. Cross-predict methods are used to estimate the flow prediction quality. It
ce
consists of adjusting the CFD results to the measurements at one position taken as
Ac
reference and estimating values for another meteorological mast (met mast), which
are then compared against the observations. The available results of the Bolund field
SOL-16-1465, SILVA 3
experiment will be used to check the quality of the proposed approach. The dataset
According to Fallo [13], speed-up can be defined as the ratio between the wind
speed in one point of the map and the speed in another particular reference point,
d
ite
where measurements should be known. Besides, Meng [6] describes the use of speed-
ed
up ratios as the relative change of a given variable that is under evaluation, instead of
py
their actual values. This method has been employed for wind estimation analysis in
Co
relevant previous works, such as [8], [14-16].
Ongoing tools usually adjust the wind field results to each instrument height and
ot
the speed-ups are also based on a single height [17-19]. On the other hand, in this
tN
work, the calculated flow is adjusted along all met mast measurement heights through a
ip
Least Squares Method. Thus, the goal is the vertical wind profile instead of the velocity
r
sc
at single heights.
nu
In the present study, a procedure to fit modeled wind profiles to the measured
Ma
ones will be presented taking into account results of CFD simulations performed with
the commercial software WindSim over the area of the Bolund hill benchmark study.
ed
the open literature are Clerc et. al [20] and Silva et. al [21]. The former presented a
ce
method to determine the best locations for installing met masts in a wind farm,
Ac
optimizing their weights for using in wind flow models. They also employed the speed-
up methodology, but the calculations considered only the hub height. They did not
mention any treatment for vertical wind profiles. Silva et. al [21] proposed a procedure
SOL-16-1465, SILVA 4
to correct LIDAR profiles using simulated and experimental data. However, they worked
with two positions only, and the correction factor was calculated height by height,
showing that the correction was not performed using a mean fitted profile as it will be
proposed herein. In the present study, a greater number of points are applied for cross-
d
ite
predictions, besides the connection to the flow over a complex terrain, allowing a
ed
deeper analysis. These distinctive features clarify the contribution of this work, when
py
compared with previous studies.
Co
2. DATA AND METHODS
Some years ago, the Bolund hill was chosen as object of study of a project
sc
microscale models. Located near Ris Campus - National Laboratory for Sustainable
Ma
Energy (Denmark), it is surrounded by water, connected to the main land only through a
narrow isthmus (Fig. 1a). Its dimensions are 130 x 75 m and the maximum height is only
ed
12 m on its flat top [22]. Despite being small, Bolund represents a typical complex
pt
terrain, with steep slopes and almost vertical cliffs reaching the sea (Fig. 1b). These
ce
geometries induce complex effects in the flow around the hill, including recirculation
Ac
The regions topography was scanned using an airborne laser terrain mapper in a
very high resolution. The orography grid file has been shared to the scientific
SOL-16-1465, SILVA 5
terrain model (DTM), with a 0.25 m resolution, extends along 290 x 250 m. It covers the
d
ite
masts at different levels (up to 16 m height), including in the sea, performed
ed
measurements providing a large data base for model validations [22, 24]. The
py
positioning of the masts is shown in Fig. 2. The predominant wind direction in Bolund is
Co
West - SouthWest [22, 25]. M0 was the only one installed over the water, in order to
accommodate the westerly undisturbed winds originating from the fjord. M1 and M7
ot
were placed right at the base of the hill, on the short terrain extension between the
tN
water and the main escarpment. M2, M3 and M6 were on the top of the hill. M4, M5
ip
and M8 were also positioned on the lower part, but around the smooth inclinations. M9
r
sc
was located at the beach of the main land and receives disturbed winds from East. In
nu
the current study, M9 has been disregarded in order to keep only water boundaries. For
Ma
the numerical simulations, the roughness class was set constant at all land grid points,
deviations) for three wind speed components, turbulent kinetic energy and its three
ce
components of variances. They are available for each anemometer and only four
Ac
Both measurements and the digital terrain model data are available in the
SOL-16-1465, SILVA 6
equations (PDE's) that represent a physical flow field. Through a discretization scheme,
d
ite
number of nodal points. The PDE's are then integrated over the computational grid and
ed
transformed into a system of algebraic equations, that is solved iteratively until a
py
converged solution is obtained according to pre-established error criteria [27].
Co
CFD RANS equations solvers had an extensive development over the 70s and
80s decades and is considered for some authors as the scientifically most promising
ot
tool for computational simulation and even the state-of-the-art in wind flow modeling
tN
[15, 28]. In wind energy, CFD RANS solvers have shown to be suitable for high resolution
ip
and was developed in Norway in the 90s. Considering steady-state and incompressible
Ma
flow, by default, the RANS equations are discretized and integrated on the
Collocated Velocity solver (GCV) has been employed to solve the velocity-pressure
pt
coupling.
ce
respectively.
SOL-16-1465, SILVA 7
=0 (1)
= + + (2)
d
ite
Eq. 1 represents the mass balance inside the control volume, where U denotes
ed
the velocity vector with its components (u, v, w). Eq. 2 expresses the balance of
py
transport of fluid with source and sink terms due to pressure and viscosity. and p are
Co
the constant air density and pressure, and is the kinematic viscosity. The turbulent
Reynolds stresses relate to the mean velocity variables through the turbulent viscosity
ot
t:
tN
ip
= + + (3)
r
sc
nu
The turbulence must be parameterized through some closure model. The most
Ma
common (and adopted in this work) is the standard k-, represented in the following
equations:
ed
pt
= (4)
ce
Ac
where k and are the turbulent kinetic energy and its dissipation rate, respectively.
! $%
= # (+) * (5)
&'
SOL-16-1465, SILVA 8
! $%
= # (+, ) , (6)
&+
d
ite
There are more computational intensive turbulence models that aim to capture
ed
finer specific characteristics of a turbulent flow. The LES approach is an example, which
py
is very suitable for unsteady flow simulations [7, 30, 31]. However, it requires very
Co
higher grid resolution, demanding extra-large computational resources, which,
nowadays, limits its utilization in wind power engineering applications - wind farm
ot
design requires a large number of simulations (at least 12 sectors and extensive surface
tN
areas).
ip
For the simulations performed in WindSim, the area of the geometric domain
r
sc
was extended in relation to the provided DTM to achieve a more accurate imposition of
nu
the boundary conditions. The extended area comprises 483 x 337 m, centered at the
Ma
central coordinate of the hill. Only water grid points were added, as can be seen
highlighted in Fig. 1a. Additional grid configurations will be further discussed. The
ed
Roughness length (z0) equal to 0.015 m over land and 0.0003 m over water (project
ce
recommendation [15]);
Ac
No grid orthogonalization;
SOL-16-1465, SILVA 9
The procedure hereafter presented was used to adjust the wind profiles using
d
the known information about the mean wind observed in Bolund met masts. It is
ite
applied over the CFD simulated wind profiles taking into account each of the four
ed
considered sectors.
py
The adjustments were made only for the calculated and measured values at a
Co
same height (level). When the anemometers levels were not in the CFD grid,
sum of square errors (f) is defined according to Eq. 7. It correlates the differences
r ip
between the simulated and the measured wind speeds (VCFD and Vobs) in all available
sc
In order to minimize the quantity in f to best fit the curve (in this case the vertical
ce
wind profile), we need to find the value of A that minimizes f, what is solved by taking
Ac
the derivative of f with respect to A equal to zero (Eq. 8). In sequence, A must be
isolated to obtain the least square estimator and perform predictions to other positions
SOL-16-1465, SILVA 10
d
ite
J
KFG =BCD =>?@ !
4= (10)
ed
J
KFG#=>?@ (
py
Co
This last parameter, termed A factor, symbolizes the accuracy of the mean
wind profile found by the model in comparison to the measured one, constructed from
ot
the data collected by the instruments at the same mast. The value 1 means total
tN
agreement. It was here calculated minimizing the differences between predicted and
ip
measured wind speeds taking into account all the heights of anemometers installed in a
r
sc
met mast, while other authors ([18-20], [32, 33], for example) calculate it using only a
nu
Next, the masts were taken, one by one, as the reference position to extrapolate
(or estimate) wind speeds to other positions of the domain, as if it were the only one.
ed
(In our specific case, these other positions are exactly the other masts, where the
pt
observed wind speeds are known, allowing a later cross comparison. But, in reality, any
ce
point could be estimated.) The wind speed estimation in a new position (Vestx) is
Ac
obtained by multiplying VCFD of the new point and the correction factor (A) of the
SOL-16-1465, SILVA 11
where y is the mast taken as reference and x is the new point where the wind speed will
be estimated.
d
ite
Finally, to validate the results, this estimation (Vest) is compared against the
ed
observed value (Vobs) on that mast. Root Mean Square Error (RMSE) was determined as
py
statistical metric to evaluate the quality of the wind profile corrections, considering all h
Co
heights, as in Eq. 12.
ot
RMSE = Z[ [^: /VMNO V\]N 8 (12)
tN
r ip
Mean RMSE were calculated among the estimations made by each reference
sc
mast to evaluate its quality as predictor. Similarly, prediction quality of the wind profile
nu
at a given point is indicated by the mean RMSE calculated among all references used to
Ma
estimate it.
SOL-16-1465, SILVA 12
3. RESULTS
The grid configurations of the CFD simulation herein used to implement the
d
ite
in a previous work [34].
ed
Regarding the grid independence for horizontal and vertical resolution, six
py
different horizontal grid spacing were tested and no significant changes were observed
Co
when using 3.0 m or less. To avoid large cell distortions, high vertical resolutions were
also tested. 51 vertical levels were defined with 15 of them concentrated inside the first
ot
50 meters height.
tN
models of Atmospheric Boundary Layer (ABL): the boundary layer height (Hbl) and the
r
sc
domain height. Since information about ABL are not available and in order to avoid
nu
interference on CFD results, these parameters were also tested. Based on the
Ma
differences among results in both cases, Hbl = 80 m and 400 m for domain height have
been chosen and adopted in the subsequent studies. This is coherent with the local
ed
characteristics of boundary layer developing over the sea, which is a flat surface with
pt
very low roughness. This value for domain height is several times greater than the
ce
maximum height of the Bolund hill (12 m). These values were also used in Bechmann et
Ac
Therefore, the following grid setup was adopted as the most suitable to the
SOL-16-1465, SILVA 13
Hbl = 80 m;
51 vertical levels.
d
ite
3.2 Speed-up Analyses
ed
Results of cross-predict speed-up studies between each pair of the 9 mast
py
positions located inside the simulation domain are herein presented. The numerical
Co
simulation utilized for that was carried out with the aforementioned model settings.
ot
Before performing cross-predictions, the adjustment of the simulated winds to
tN
the measured ones at the reference position is verified and presented in Fig. 3. Vertical
wind profiles are shown for all 9 positions including the observed, the simulated and the
r ip
curve that best fit them, obtained through the A factor, which minimizes their
sc
differences at each position. The correction was applied using only the corresponding
nu
observed levels.
Ma
By using the least squares fitting to the vertical wind profile its possible to
ed
observe that the adjusted curves approximate very well to the observed profiles in most
cases. When the agreement is good, we can infer that the CFD problem setup was
pt
ce
suitable (flow modeling, turbulence model, orography, roughness, grid and boundary
Ac
conditions). In this case, CFD results are a good predictor for the vertical wind shear. A
good estimation for the wind shear is very important to analyze economic feasibility of
taller wind generator towers. When the hub height is increased, more kinetic wind
SOL-16-1465, SILVA 14
energy is available, however the complete wind turbine becomes more expensive.
Thereby, the additional captured wind energy should pay the extra cost.
The vertical profile adjustments were not so good for masts ID M2, M6 and M8,
for which the adjusted curves were not able to correct the vertical wind shear. These
d
ite
cases presented large deviations in vertical wind profile between simulation and
ed
observation. Considering that these results are for incoming winds from 270, M2 and
py
M6 were located immediately above the steep wall, and M8 on the bottom of the
Co
opposite side, downstream of the hill (see Fig. 2). These positions are in areas subjected
to occurrence of recirculation phenomenon when the winds come from West directions,
ot
as detected by Bechmann et al. [25] in their simulations. Also Prospathopoulos et al. [5]
tN
found less accurate simulation results in the masts located downstream the
ip
escarpment. They stated that incoming winds from these directions are more
r
sc
complicated to solve because the flow reaches the cliff in an almost vertical inclination,
nu
formation.
Next, Fig. 4 exemplifies the ability of using these first adjustments at each
ed
reference position (from here on called the predictor) to estimate the wind profiles in all
pt
other positions (from here on called the predicted). Face to the unfeasibility of
ce
presenting all set of figures, i.e., for each mast as reference, only one is here illustrated.
Ac
M7 is shown as the reference mast considering the same direction as above. It meets
predominant wind directions, which flow transports the effects downwind. One can
SOL-16-1465, SILVA 15
notice that the adjustment has improved the CFD estimations in most but not all cases.
The corrections for M5 and M8, for example, got further from the observed curve than
the uncorrected simulated profile, itself. It shows that the CFD model has not been able
to calculate accurately the wind speed variation between two masts, i.e., the speed-up
d
ite
variation.
ed
Table 1 shows the whole set of cross-prediction results according to the available
py
observed dataset. Statistics of individual RMSE (for all possible cross-correlations
Co
between predictor and predicted masts), A factor and mean RMSEs are presented.
In addition, an analysis has been performed to check the role played for each
ot
mast when used as reference position, in order to identify if one of them has prevailed
tN
as best predictor when compared to the others. This is interesting face to the need of
ip
choosing the local for positioning a single anemometric tower in a region of interest for
r
sc
a feasibility study. The analysis must take into account the characteristics of the flow at
nu
As can be seen in Table 1, in relation to the mean RMSE of predictors (last line of
each package), M2 presents the lowest errors for the Southwestern cases and M8 for
ed
the Eastern one. In contrast, this same M8 is the worst predictor for Southwesterly
pt
winds. This behavior is coherent with the masts positions since M8 is at the East tip of
ce
Bolund, on the lower part of the terrain. When the incoming winds are from East, the
Ac
flow at that position is not disturbed by the orography effects of the hill. But when the
winds come from Southwest, M8 is on the lee side of the hill and the extrapolation to
SOL-16-1465, SILVA 16
Based on results of this study, the ability of being a good predictor depends on
the flow development downstream of that position for a given wind direction. M2 and
M6 are at the top of the hill and the Western flow continues downstream over a gentler
orography. As well as the Easterly flow passing through M8 goes over the hill through
d
ite
relatively smooth inclinations, without steep walls.
ed
On the other hand, although M2 is a good predictor position, it is very difficult to
py
be predicted, as well as M6, as one can see in the table. These two masts have been
Co
verified to be the most difficult positions to estimate wind profiles, presenting the most
of the highest prediction errors (individual RMSE), no matter which is the predictor
ot
mast. The mean RMSE of predicted in the rightmost column facilitates this analysis.
tN
Thus, if the flow characteristics are more complex, the profile correction will be
ip
worse. As already discussed at the beginning of the present section, this result may be a
r
sc
limitation of the wind flow model utilized (CFD, turbulence modeling, grid refinement) in
nu
results.
Lastly, results found here did not show direct correlation related to the distance
ed
over which the prediction is made. The area is small enough for that the topographical
pt
4. CONCLUDING REMARKS
In this work, CFD simulated vertical wind profiles were adjusted to observed
ones through a least squares fitting. These adjusted wind flow results were used to
SOL-16-1465, SILVA 17
procedure. The methodology is useful to improve CFD results, which just take into
account the terrain effects on simulating the wind fields. One advantage of this process
is that it considers both the mean profile and the wind shear, and can provide more
d
ite
accurate extrapolations than using individual heights. Besides, the adjustment done
ed
taking into account measurements from more than one height can reduce the
py
experimental uncertainty.
Co
Results showed that the performance of the wind profile estimations is strongly
affected positions related to the inflow direction present the less accurate wind profile
ip
estimations. In the Bolund hill, particularly, this is critical for incoming flow from
r
sc
SouthWest due to the perturbation effect that occurs when the airflow overcomes the
nu
terrain escarpment.
Ma
Face to the choice for installing one anemometric tower as reference position for
after the main local characteristics of topographical changes. This is directly dependent
pt
on the predominant wind direction at the considered site. So, in case of a wide spread
ce
of the wind directions, more than one tower may be necessary for a better adjustment
Ac
When a good CFD adjustment is obtained for the reference mast, and the results
for the test mast are in good agreement with the observed ones, one can conclude that
SOL-16-1465, SILVA 18
the CFD model was able to calculate accurately the wind speed variation between two
d
ite
ed
py
Co
ot
tN
r ip
sc
nu
Ma
ed
pt
ce
Ac
SOL-16-1465, SILVA 19
NOMENCLATURE
d
CFD Computational Fluid Dynamics
ite
DTM Digital Terrain Model
ed
Hbl Boundary Layer Height
py
PDE Partial Differential Equation
Co
RANS Reynolds Average Navier-Stokes
ot
RMSE Root Mean Square Error
tN
h Measurement heights
sc
p Pressure
Ma
U Velocity vector
pt
z0 Roughness length
SOL-16-1465, SILVA 20
Kinematic viscosity
t Turbulent viscosity
d
ite
ed
py
Co
ot
tN
r ip
sc
nu
Ma
ed
pt
ce
Ac
SOL-16-1465, SILVA 21
REFERENCES
[1] Rabl, A., and Spadaro, J. V., 2016, External Costs of Energy: How Much Is Clean
Energy Worth?, Journal of Solar Energy Engineering, 138(4), pp. 1-8. DOI:
10.1115/1.4033596
[2] Zajaczkowski, F. J., Haupt, S. E., and Schmehl, K. J., 2011, A preliminary study of
d
assimilating numerical weather prediction data into computational fluid dynamics
ite
models for wind prediction, J. Wind Eng. Ind. Aerodynamics, 99(4), pp. 320-329. DOI:
10.1016/j.jweia.2011.01.023
ed
[3] Jafari, S., Chokani, N., and Abhari, R. S., 2012, An Immersed Boundary Method for
py
Simulation of Wind Flow Over Complex Terrain, Journal of Solar Energy Engineering,
134(1), pp. 1-12. DOI: 10.1115/1.4004899
Co
[4] Wang, Y., Wang, J., and Wei, X., 2015, A hybrid wind speed forecasting model
based on phase space reconstruction theory and Markov model: A case study of wind
ot
farms in northwest China, Energy, 91, pp. 556-572. DOI: 10.1016/j.energy.2015.08.039
tN
[5] Prospathopoulos, J. M., Politis, E. S., and Chaviaropoulos, P. K., 2012, Application of
a 3D RANS solver on the complex hill of Bolund and assessment of the wind flow
predictions, J. Wind Eng. Ind. Aerodynamics, 107-108, pp. 149-159. DOI:
ip
10.1016/j.jweia.2012.04.011
r
sc
[6] Meng, P. S., 2004, CFD Modelling of Wind Flow Over Terrain, Ph.D. thesis,
Nottingham Univ., Nottingham, UK.
nu
[7] Ayotte, K. W., 2008, Computational modelling for wind energy assessment, J.
Ma
[8] Rasouli, A., and Hangan, H., 2013, Microscale Computational Fluid Dynamics
Simulation for Wind Mapping Over Complex Topographic Terrains, Journal of Solar
ed
[9] Albani, A., and Ibrahim, M. Z., 2014, An Assessment of Wind Energy Potential for
Selected Sites in Malaysia Using Feed-in Tariff Criteria, Wind Engineering, 38(3), pp.
ce
[10] Blocken, B., Hout, A., Dekker, J., and Weiler, O., 2015, CFD simulation of wind flow
over natural complex terrain: Case study with validation by field measurements for Ria
de Ferrol, Galicia, Spain, J. Wind Eng. Ind. Aerodynamics, 147, pp. 43-57. DOI:
10.1016/j.jweia.2015.09.007
SOL-16-1465, SILVA 22
[11] Simes, T., and Estanqueiro, A., 2016, A new methodology for urban wind
resource assessment, Renewable Energy, 89, pp. 598-605. DOI:
10.1016/j.renene.2015.12.008
[12] Ayotte, K. W., Davi, R. J., and Coppin, P. A., 2001, A simple temporal and spatial
analysis of flow in complex terrain in the context of wind energy modelling, Boundary-
Layer Meteorol., 98, pp. 275-295. DOI: 10.1023/A:1026583021740
d
ite
[13] Fallo, D., 2007, Wind energy resource evaluation in a site of central Italy by CFD
simulations, MSc. dissertation, Dept. Mech. Eng., Cagliari Univ., Cagliari.
ed
[14] Castro, F. A., Palma, J. M. L. M., and Lopes, A. S., 2003, Simulation of the Askervein
py
flow. Part 1: Reynolds Averaged NavierStokes equations (k turbulence model),
Boundary-Layer Meteorol., 107, pp. 501-530. DOI: 10.1023/A:1022818327584
Co
[15] Bechmann, A., Srensen, N.N., Berg, J., Mann, J., and Rthor, P.-E., 2011, The
Bolund Experiment, Part II: Blind Comparison of Microscale Flow Models, Boundary-
Layer Meteorol.,141(2), pp. 245271. DOI: 10.1007/s10546-011-9637-xot
tN
[16] Castellani, F., Astolfi, D., Burlando, M., and Terzi, L., 2015, Numerical modelling for
wind farm operational assessment in complex terrain, J. Wind Eng. Ind. Aerodynamics,
147, pp. 320-329. DOI: 10.1016/j.jweia.2015.07.016
r ip
[17] Troen, I., and Petersen, E. L., 1989, European Wind Atlas, Ris National Laboratory,
sc
[19] Meissner, C., 2015, WindSim Getting Started: WindSim 7, Technical Report,
WindSim AS, Tnsberg.
ed
[20] Clerc, A., Anderson, M., Stuart, P., Habenicht, G., 2012, A systematic method for
quantifying wind flow modelling uncertainty in wind resource assessment, J. Wind Eng.
pt
[21] Silva, J., Marques da Silva, F., Couto, A., and Estanqueiro, A., 2015, A method to
correct the flow distortion of offshore wind data using CFD simulation and experimental
Ac
wind tunnel tests. J. Wind Eng. Ind. Aerodynamics, 140, pp. 87-94, DOI:
10.1016/j.jweia.2015.01.017
[22] Bechmann, A., Berg, J., Courtney, M. S., Hans, E., Mann, J., and Srensen, N. N.,
2009, The Bolund Experiment: Overview and Background, Technical Report No. Ris-
R-1658, Wind Energy Division, Ris DTU, Roskilde, Denmark.
SOL-16-1465, SILVA 23
[24] Berg, J., Mann, J., Bechmann, A., Courtney, M. S., and Jrgensen, H. E., 2011, The
Bolund Experiment, Part I: Flow Over a Steep, Three-Dimensional Hill, Boundary-Layer
Meteorol., 141(2), pp. 219243. DOI: 10.1007/s10546-011-9636-y
d
[25] Bechmann, A., Johansen, J., and Srensen, N. N., 2007, The Bolund Experiment
ite
Design of measurement campaign using CFD, Technical Report No. Ris-R-1623, Wind
Energy Division, Ris DTU, Roskilde, Denmark.
ed
[26] Bolund Blind Comparison, 2016,
py
<http://www.bolund.vindenergi.dtu.dk/blind_comparison>, [accessed 2015 Oct 8].
Co
[27] Versteeg, H. K. E., and Malalasekera, W., 1995, An Introduction to Computational
Fluid Dynamics: the Finite Volume Method, Longman Scientific & Technical, New York,
USA. ISBN: 0-582-21884-5
ot
[28] Theodoropoulos, P., and Deligiorgis, N., 2009, WindSim CFD model validation in a
tN
mixed coastal & mountainous region with complex terrain, Proc. European Wind
Energy Conference, Marseille.
ip
[29] Gravdahl, A. R., 1998, Meso scale modeling with a Reynolds Averaged Navier-
r
Stokes Solver: Assessment of wind resources along the Norwegian coast, Proc. 31st
sc
[30] Chow, F., Street, R., 2009 Evaluation of turbulence closure models for large-eddy
simulation over complex terrain: flow over Askervein Hill, J Appl Meteorol., 48(5), pp.
Ma
1050-1065. DOI:10.1175/2008JAMC1862.1
[31] Lee, S., Churchfield, M. J., Moriarty, P. J., Michalakes, J., 2013, A Numerical Study
ed
[32] Moreno, P., Gravdahl, A., Romero, M., 2003, Wind flow over complex terrain:
ce
application of linear and CFD models, Proc. European Wind Energy Conference and
Exhibition, Madrid, pp. 16-19.
Ac
[33] Gonzlez-Longatt, F., Gonzlez, J. S., Payn, M. B., Santos, J. M. R., 2014, Wind-
resource atlas of Venezuela based on on-site anemometry observation, Renewable and
Sustainable Energy Reviews, 39, pp. 898-911, DOI: 10.1016/j.rser.2014.07.172
SOL-16-1465, SILVA 24
[34] Silva, A. F. G., and Zaparoli, E. L., 2014, The use of CFD as a tool for studying the
wind energy in complex terrain, Proc. 18th Congresso Brasileiro de Meteorologia,
Recife.
d
ite
ed
py
Co
ot
tN
r ip
sc
nu
Ma
ed
pt
ce
Ac
SOL-16-1465, SILVA 25
Fig. 1 Bolund hill view (a) Google Earth; (b) Bolund Background [22]
d
Fig. 3 Observed, simulated and adjusted vertical wind profiles, all for 270
ite
Fig. 4 Observed, simulated and estimated vertical wind profiles, with
ed
correction through M7 as reference to all predicted positions, all for 270
py
Co
ot
tN
r ip
sc
nu
Ma
ed
pt
ce
Ac
SOL-16-1465, SILVA 26
d
ite
ed
py
Co
ot
tN
r ip
sc
nu
Ma
ed
pt
ce
Ac
SOL-16-1465, SILVA 27
a b
d
ite
ed
Fig. 1. Bolund hill view (a) Google Earth; (b) Bolund Background [23]
py
Co
ot
tN
r ip
sc
nu
Ma
ed
pt
ce
Ac
SOL-16-1465, SILVA 28
d
ite
ed
py
Fig. 2. Positioning of the masts (Adapted from Prospathopoulos et al. [5])
Co
ot
tN
r ip
sc
nu
Ma
ed
pt
ce
Ac
SOL-16-1465, SILVA 29
d
ite
ed
py
Co
ot
tN
Fig. 3. Observed, simulated and adjusted vertical wind profiles, all for 270
r ip
sc
nu
Ma
ed
pt
ce
Ac
SOL-16-1465, SILVA 30
d
ite
ed
py
Co
ot
tN
Fig. 4. Observed, simulated and estimated vertical wind profiles, with correction through
ip
SOL-16-1465, SILVA 31
Predictors
M0 M1 M2 M3 M4 M5 M6 M7 M8
Mean
A factor 1.049 1.072 0.847 0.946 1.001 0.966 0.817 0.725 0.942
RMSE
d
M0 - 0.393 1.688 0.908 0.520 0.761 1.926 2.669 0.941 1.226
ite
M1 0.306 - 1.541 0.890 0.547 0.764 1.737 2.351 0.918 1.132
M2 1.962 2.170 - 1.099 1.548 1.256 0.666 1.275 1.066 1.380
ed
M3 0.913 1.103 0.884 - 0.545 0.332 1.122 1.881 0.289 0.883
Predicted 90
py
M4 0.396 0.557 1.154 0.444 - 0.316 1.371 2.046 0.474 0.845
M5 0.675 0.849 0.481 0.261 0.344 - 1.167 1.869 0.282 0.741
Co
M6 2.469 2.469 0.590 1.324 1.815 1.499 - 1.013 1.287 1.558
M7 1.359 1.359 2.989 3.159 1.095 0.966 0.461 - 0.879 1.533
M8 0.720 0.859 0.648 0.262 0.454 0.301 0.821 1.379
ot - 0.680
Mean
RMSE 1.100 1.220 1.247 1.043 0.858 0.774 1.159 1.810 0.767
tN
Mean
A factor 1.095 1.078 0.984 1.222 1.067 0.830 0.775 0.977 0.534
ip
RMSE
M0 - 0.161 0.913 1.038 0.246 2.167 2.612 0.966 4.575 1.585
r
sc
Mean
RMSE 2.150 2.107 1.661 2.685 1.907 2.325 2.092 2.070 3.610
ce
Ac
SOL-16-1465, SILVA 32
Table 1. Continuation
Mean
A factor 1.129 1.127 1.035 1.218 1.217 - 0.946 1.043 0.641
RMSE
M0 - 0.090 0.769 0.734 0.726 - 1.496 0.710 3.984 1.215
M1 0.083 - 0.518 0.519 0.513 - 1.015 0.478 2.718 0.835
d
M2 3.112 3.108 - 3.354 3.350 - 3.104 3.022 4.356 3.344
ite
M3 0.608 0.623 1.242 - 0.046 - 1.848 1.192 3.917 1.354
Predicted 255
ed
M5 - - - - - - - - - -
py
M6 4.372 4.366 4.183 4.656 4.652 - - 4.194 4.783 4.458
M7 0.414 0.404 0.035 0.843 0.838 - 0.464 - 1.929 0.704
Co
M8 2.532 2.523 2.149 2.913 2.908 - 1.805 2.178 - 2.430
Mean
RMSE 1.738 2.012 1.572 2.206 2.051 - 1.647 2.058 3.451
ot
Mean
tN
A factor 1.375 1.346 1.153 1.527 1.171 0.853 1.389 1.242 0.673
RMSE
M0 - 0.248 1.810 1.234 1.660 4.245 0.135 1.085 5.715 2.017
ip
Mean
2.278 2.499 1.891 3.202 2.003 3.365 2.292 2.381 4.300
RMSE
pt
ce
Ac
SOL-16-1465, SILVA 33