Wind energy yield analysis at large scale

MEC583: Projects on Solar and Wind Energy:

Resource and Analysis

Submitted By Submitted to
Anuj Keshav Prof. Sylvain Cros
Davoud Shalforoushan

1
 Table of Content
Motivation…………………………………………………………….3
Objectives……………………………………………………………..3
Methodology…………………………………………………………..4
Results……………………………………………………………..…..9
Conclusion……………………………………………………………17
References………………………………………………………….....18
Appendix….…………………………………………………………..18

2
 1.Motivation
This project aims to assess the financial feasibility of a wind farm located in a specific region of France. In order
to determine the financial viability of the wind farm, several variables need to be considered. These include a
detailed analysis of seasonal wind speeds at the site, as well as the wind potential power.

To calculate the wind potential power, data on wind speeds at the location must be collected and analysed. This
includes not only the typical wind speeds that can be expected, but also the frequency of extreme wind events
that can occur. It is also important to identify the wind resources at that location, as this can have a significant
impact on the overall power output of the wind farm. The wind potential power calculated from this data is a
critical factor in determining the overall power output of the wind farm, which in turn directly affects the revenue
that can be generated from the sale of electricity. This revenue stream is what determines the financial viability
of the project.

To accomplish the goals of this project, a variety of data analysis techniques will be utilized. This will include
techniques such as interpolation, compiling data into pandas data frames, time series analysis, curve fitting, and
power time series analysis. These techniques will be used to ensure that the wind data is accurately analysed
and that the resulting wind potential power calculations are as precise as possible.

2.Objectives
The objective of this project is to evaluate the financial feasibility of wind energy production in France
comprehensively. The evaluation process will involve analysing several critical factors that influence wind
energy production, such as seasonal wind speed patterns, frequency of extreme events, wind potential power
generated by different wind farms, surface wind reanalysis, and power curves from wind turbines. Each of these
factors will be thoroughly examined to provide an accurate estimate of the yearly average power production of
wind farms at different locations, and a quantification of the risk of over- or under-estimating wind energy yield.

The ultimate goal of this project is to help maximize the financial return for all stakeholders involved in wind
energy production in France. To achieve this goal, it is necessary to consider various variables that impact the
analysis of the financial viability of wind energy production. These variables include the factors mentioned
earlier, as well as other factors such as capital and operating costs, regulatory policies, and market demand for
wind energy. To count the factors which effect the analysis of the financial viability of wind energy production
in France, we will encounter the following variables:

1. Seasonal wind speed patterns and the frequency of extreme events. (percentile
50, 75 and 90, extreme events frequency)
2. Wind potential power computing from wind data analysis for wind farms
3. Evaluate the financial return of a wind farms portfolio in France.
4. Quantification of the risk of wind energy yield over- or under-estimation.
3
 5. The yearly average power production of a wind farms on all the locations
6. With the goal of maximizing the financial return for the stakeholders.

3.Methodology
• Acquiring the data from ERA5
ERA5 is the fifth generation ECMWF atmospheric reanalysis of the global climate covering the period from
January 1950 to present. ERA5 is produced by the Copernicus Climate Change Service (C3S) at ECMWF.

ERA5 provides hourly estimates of a large number of atmospheric, land and oceanic climate variables. The data
cover the Earth on a 30km grid and resolve the atmosphere using 137 levels from the surface up to a height of
80km. ERA5 includes information about uncertainties for all variables at reduced spatial and temporal resolu-
tions. If the data for the wind is plotted in the graph it will look like the Fig (1).

Fig(1) Mean wind speed estimates at 100 m, calculated based on data obtained... | Download Scientific Diagram (researchgate.net)

The data for winds, temperature and weather is usually stored in the NetCFD4 file format. The netCDF4 format
supports data stored in an array format. Arrays are used to store raster spatial data (terrain layers, gridded
temperature data, etc) and point based time series data (for example temperature for a single location over 10
years). Climate data typically have three dimensions—x and y values representing latitude and longitude
location for a point or a grid cell location on the earth’s surface and time. In our project the data is stored with
one variable u100 or v100 and the parameters such as latitude, longitude and timestep. We are using the u100
and v100 variable as most of the turbine hub starts at 50-60 m and goes up to 130m and this value for wind will
give us close to true value of the wind.

4
 • Wind farm locations
Our study mainly focuses on the wind farms in France, the list of farms and their coordinated are mentioned the
table below. Some of the farms are near coast and some are in the landlocked position geographically. The data
is shown below in Table 1 and Fig (2).

Location
City Lati- Longi-
tude tude
Longèves 46.14 -0.58
Maisonnais 45.42 0.4
Tourny 49.1 1.37
Riaucourt 48.1 5.11
Smermesnil 49.84 1.49
Le Buret 47.54 -0.29
Puchot 49.41 1.59
Tillières 47.9 -1.8
Plaine Conli-
48.7 0
noise
Merisier 47.25 -1.24
Malaucourt 48.5 6.2
Daméraucourt 49.41 1.55

Table (1), List of 12 wind farms. (2). The position of 12 farm stations.

• Wind Speed data for each location

There are several ways to extract data from the NetCFD4 type file. The libraries such as xarrays and netcfd4 is
used to extract the desired data. The data stored in the netCFD4 file has a lot of information on variables and
latitude and longitude, one such parameters is called degree-step, it simply tells us about the difference between
two neighbouring latitude or longitude location. It is highly likely that the data we need for particular location
doesn’t coincide with the data point location (latitude, longitude) of the stored data. Also, during the extraction
of the data you might come across error such as ‘Memory error’, which indicate that the data size is too large
for computer memory to compute, in that case the datetime library can be used to slice the data based on the
timestep e.g., ‘VAR = VAR.sel(time=slice(time_start,time_end))’here VAR is the extracted variable data and
time_start and time_end is defined for the start and end time.

To extract the wind speed data for all the cities, the data for each variable u100 and v100 is extracted based on
the neighbouring locations of the desired location, the netCFD4 library was used to extract the wind speed data.
The bilinear interpolation was used to get the data for the desired location. For this application, linear
interpolation works well since the wind speed data exhibits smooth spatial and temporal variation. The wind
speed numbers are interpolated for the place of interest using bilinear interpolation, a sort of linear interpolation
that takes four nearby grid points into account while calculating the value at an arbitrary point. This approach,
which is suitable for this kind of data, presupposes a linear relationship between the values at adjacent places.

5
 To interpolate the data, we first need to find the nearest points to the desired location that are available in the
dataset. To get the data for neighbouring location the degree step of 0.25 degree was used, we can extract the
wind speed data for each of these points from the netCDF4 file. The mathematical expression for the bilinear
interpolation is shown below. This mathematical expression can be converted into function the python with
inputs like coordinated and the respective wind speeds.

Fig (3). c++ - Bilinear interpolation artifacts - Stack Overflow

Here, f(x,y) = desired wind speed , x= desired location (longitude) , y = desired location (latitude), l and k are
the neighbouring locations, in this formula the degree step is 1 but in the computation 0.25 is used. After the
interpolation the data is stored in the panda data frame and the date column is added to it. All these steps are
done only for the u100 variable which signifies the wind speeds from west to east, now the same steps are to be
done for the v100 variable which gives the value of wind speeds from north to south.

Fig (4). Component of wind speed

After acquiring all the data for all the locations for all the variables such as u100 and v100. To get the total
wind speed for all the location we will square and add the data. The expression is as follows: -
6
 Total wind speed = √𝑢1002 + 𝑣1002

• Computation and analysis the Power production.

The power production is computed by interpolation from the reference data. The graph will show the accuracy
if the interpolation is correct for further analysis.

Fig (5). Reference Power data

Fig (6). Interpolated Power data

The reference data for the power production data is taken from actual wind farm. We can observe that the
interpolation of data is very accurate, and we can make the further analysis of the power production based on
this data. The interpolation is done using the SciPy library in python and function used for interpolation is
‘scipy.interpolate’ which is an inbuilt linear interpolation function.

7
 • Time series analysis for wind speed and power production
Time series analysis is a technique in statistics that deals with time series data and trend analysis. Time series
data follows periodic time intervals that have been measured in regular time intervals or have been collected
time intervals. In other words, a time series is simply a series of data points ordered in time, and time series
analysis is the process of making sense of this data.

The wind speed data is analysed using this technique, the matplotlib library is used to plot the data for 2 years.
Similarly, the power production data can be analysed with this to get a concrete idea.

• P50, P75 & P90 and normal distribution plot

To assess the wind energy yield potential of a site, the model is run using best available data and methods. The
result of the modelling is the P50, P75 and P90 estimate, or in other words, the “best estimate”. P50 is essentially
a statistical level of confidence suggesting that we expect that the predicted wind resource/energy yield may be
exceeded with 50% probability.

Fig (10). Probability distribution

P50, P75, and P90 are all terms used to describe the probability of availability for a given resource or
production.

• P50 means that there is a 50% probability that the resource or production will be available for use. In
other words, the resource or production will be available about half the time.
• P75 means that there is a 75% probability that the resource or production will be available for use. In
other words, the resource or production will be available about 25% of the time.
• P90 means that there is a 90% probability that the resource or production will be available for use. In
other words, the resource or production will be available about 10% of the time.

8
 These probabilities are often used in the context of planning for resource or production usage, as they can give
us an idea of how much of a given resource or production, they can also be used for setting benchmarks or
targets for power production, as well as for identifying potential issues or opportunities for improvement.

Calculating percentiles and statistical parameters of the wind speed distribution is useful for understanding the
characteristics of the wind speed data. Also, the normal distribution is used to fit a probability distribution to
the wind speed data for each location. Also, we can obtain the mean and standard deviation of the distribution,
which can be used to describe the shape and spread of the data, we used the scipy.stats.norm function to fit a
normal distribution to the wind speed data. Here, we analyse the wind speed distributions for each location. We
define a function that takes a Data frame and a column name as inputs, and returns the 50th, 75th, and 90th
percentiles of that column. We then loop over each location in our list of the cities, we read the wind speed data
for that location, and calculate the percentiles using our function. We also calculate the mean and standard
deviation of the wind speed data.

To visualize the wind speed distribution, we use a histogram with a normal distribution curve fitted over it using
sns.distplot. We add vertical lines to mark the 50th, 75th, and 90th percentiles using ax.axvline, and add a legend
to the plot.

4.Results
Analysis of the Wind Farm
In this analysis we will be working with to extreme location in France. The first location is Longèves, and the
second location is Malaucourt. Due to following reasons, we have decided to analyse these two locations only.

• Both locations are at the extreme of France i.e., Longèves is in the west France and Malaucourt is in
east France, this will give the idea of wind production in France when going across country
longitudinally.
• Longèves is near the shore whereas Malaucourt is in landlocked position, this will give us a crucial
analysis of wind power production and effect of coast on it.

Fig (11). Location of Longèves (red) and Malaucourt (black)

9
 Longèves
The Longèves wind farm was built in the Nouvelle Aquitaine region in the west of France and was developed
and realised by the French subsidiary WKN France. With a total capacity of 9 MW, the expected annual
electricity production amounts to approx. 25 gigawatt hours.

Longèves (France) - Wind farms - Online access - The Wind Power

Latitude: 46° 14' 4.2”, Longitude: -0° 58' 15.5"

• Time Series analysis of Wind Speed over two year (2020-21).

The wind speed data for 2 years from ERA5 has been analysed using python. The results are as follows: -

Fig (12). Time series graph for Longèves

Based on the time series graph, we can make the following interpretations:

- The average wind speed at this location over two years is between 5-6 m/s.
- The fastest wind speed recorded during this time is more than 15 m/s, but it is rare.

10
 - There are days when the wind speed is close to 0 m/s, indicating very low or no wind activity.
- There are spikes in wind speed between August to October, which could be due to seasonal changes or
temporary wind patterns.
- The spike in wind speed during this period may also be related to global warming, which could be
affecting weather patterns and leading to more extreme weather conditions. Overall, the time series graph
gives us a good idea of the wind patterns at this location over a two-year period, and highlights some of
the trends and patterns that are occurring.

• P50, P75 and P90 analysis of Wind Speed

Fig (13). Probability distribution for Longèves

At first glance, it is evident that the average wind speed over two years in this location is between 5 and 7 m/s
and the values for P50, P75 and P90 are 6.56, 8.67 and 11.2 m/s respectively. This means that for 10 % of
time this location will have windspeed more than 11.2 m/s. for 50 % of time the value of wind speed is less
than 6.6 m/s.

• Time Series analysis of Power production (2020-21)

Fig (14). Time series for Power production (Longèves)

11
 The reference power curve tells us that the up to certain wind speed threshold level the wind farm does not
produce any significant power i.e. until 5 m/s. In the time series graph, there are some dips in the curve for
certain time from august to October for each year which correctly correlates to the low wind speed during that
time in the previous wind time series graph. The highest difference in the power production can be seen during
winter i.e., from December to march as wind speed changes a lot during that time, surprisingly we can see that
this wind farm produced average power most of the time compared to the maximum production which is 2400
kW.

• P50, P75 and P90 analysis of Power Production

Fig (15). P50, P75 and P90 graph for Power production (Longèves)

At first glance, it is evident that this wind farm produced no power for majority of time as we can see huge bar
for zero power production, surprisingly the bar for maximum production also high after no power production
this correlates with power production time series and possible reason could be, fast wind speed frequency which
is high but only for short times. As we move forward the value for P50 is 487 kW that means this wind farm
will produce more than 487 kW for 50 % of the time. Similarly, for P75 and P90 the values are 1167 kW and
2207 kW, this gives us an idea that this wind farm will produce power more than 1167 kW for only 25 % of
time and 2207 kW for only 10 % of the time.

Malaucourt-sur-Seille
Malaucourt-sur-Seille is a small village in the north-east of France. The village is in the Moselle department in
the Lorraine region. The wind farm in this region is developed and installed by Energieteam in 2018. This
farm has 5 wind turbine and has capacity of 12 MW.

12
 Malaucourt-sur-Seille (France) - Wind farms - Online access - The Wind Power
Latitude: 48° 50’ 41.6” Longitude: 6° 20’ 30.9”

• Time Series analysis of Wind Speed over two year (2020-21).

The wind speed data for 2 years from ERA5 has been analysed using python. The graph is as follows:

Fig (15). Time series graph for Malaucourt

Based on the graph, it appears that the wind speed in this area remains relatively constant with occasional spikes
in maximum speed. The majority of wind speeds fall within the range of 5 to 7 m/s, with infrequent spikes in
maximum speed up to 19 to 20 m/s.

The graph depicts that the wind speed in this area experiences a lot of low spikes compared to the spikes for
maximum speed. This could indicate that there are frequent periods of calm or low wind activity in the area.
Furthermore, it is noted that during the months of July, August, and September, the wind speed is at its lowest.
This suggests that there may be seasonal patterns in wind activity in this area. Overall, the graph provides
valuable information about the wind speed patterns in this area.

13
 • P50, P75 and P90 analysis of Wind Speed

Fig (16). Probability distribution for Malaucourt

From the graph, it appears that the wind speed in this area follows a normal distribution. The P50 value of
5.69696 suggests that the median wind speed in the area is approximately 5.7 m/s. This means that 50% of the
recorded wind speeds in the area are less than 5.7 m/s, while the other 50% are greater than 5.7 m/s.

The P75 value of 7.63813 indicates that 75% of the recorded wind speeds in the area are less than 7.6 m/s,
while the remaining 25% are greater than 7.6 m/s. This means that the wind speeds in this area tend to cluster
around the lower end of the speed range, with only a small percentage of recorded wind speeds reaching
higher values.

The P90 value of 9.79190 suggests that only 10% of recorded wind speeds in the area exceed 9.8 m/s. This
means that high wind speeds are relatively rare in this area, and the majority of wind speeds fall within the
lower end of the speed range.

Overall, the normal distribution and P50, P75, and P90 values provide useful insights into the wind speed
patterns in this area. Based on these values, it can be concluded that the wind speeds in this area are relatively
low and do not vary significantly from month to month, with occasional spikes in maximum speed.

• Time Series analysis of Power production (2020-21)

The power production data is plotted for two year is shown below: -

14
 Fig (17) Time series for power production (Malaucourt)

The wind farm's power output is not substantial, as evidenced by the high frequency of no production shown in
the time series graph. Although there are occasional high-power spikes, these are infrequent and produce a
relatively small amount of energy. The average power output is below 500 kW, which could explain why five
turbines were installed to compensate for low wind speeds. Wind turbines are highly dependent on wind speed,
and during the months of July, August, and September, which experience lower wind speeds, power production
is at its lowest, as expected.

• P50, P75 and P90 analysis of Power Production

Fig (18) P50, P75 and P90 graph for Malaucourt

15
 This graph of the wind farm reveals a high likelihood of no power production, as demonstrated by the significant
bar on the graph. Additionally, the median power production value is around 307 kW, indicating that half of the
time the farm generates less than or equal to 307 kW. Similarly, the 75th and 90th percentiles of power
production do not exceed 792 kW and 1649 kW, respectively, for 75% and 90% of the time. These results imply
that the wind farm produces low levels of power.

• Comparison between Wind Farms in East and West France

Fig (19) Ventusky - Wind, Rain and Temperature Maps

Comparing the wind and power production data for West and East France, there are noticeable differences in
the patterns and levels of wind speed and power production between the two regions.

In West France (Longèves), the wind speeds appear to be higher than in East France(Malaucourt) as indicated
by the maximum wind speed spikes that exceed 25 m/s. The P50 value for wind speed is 7.32104 m/s, which is
higher than the P50 value for East France. This suggests that the median wind speed in West France is
approximately 7.3 m/s, which is higher than in East France.

In terms of power production, the wind farm in West France appears to generate more power than the one in
East France. The median power production value for West France is 750 kW, which is significantly higher than
the median value for East France. Additionally, the 75th and 90th percentiles for power production in West
France are higher than those in East France. Overall, the wind and power data suggest that West France has
more favourable wind conditions for power generation than East France.

The differences in wind and power production between West and East France can be attributed to various
factors, including geography, climate, and wind turbine technology.

Geography: West France is a coastal region with a long coastline facing the Atlantic Ocean, which exposes it
to strong and consistent winds coming from the ocean. In contrast, East France is an inland region with a more
varied topography and less exposure to oceanic winds. Therefore, the wind resource in West France is more
abundant and stronger than that in East France.

16
 Climate: The climate in West France is mild and rainy, with prevailing westerly winds that generate oceanic
winds. On the other hand, the climate in East France is more continental, with colder winters and hotter
summers. This difference in climate affects wind patterns and availability, which affects the amount of power
that can be generated from wind turbines.

• Wind and Power data for all Locations

Location Wind(m/s) Power (kW) Power Production

City P5 2020
Lat Long 0 P75 P90 P50 P75 P90 (kW) 2021 (kW)
Longèves 46.14 -0.58 6.6 8.7 11.2 487.4 1167.1 2207.5 7039473.7 6424258.0
Maisonnais 45.42 0.4 5.5 7.2 8.9 263.3 659.4 1266.6 4232498.6 3977319.0
Tourny 49.1 1.37 6.1 8.2 10.2 380.6 981.2 1854.3 6247901.1 5083774.7
Riaucourt 48.1 5.11 5.8 7.7 9.6 319.0 816.6 1574.6 5233313.6 4557625.1
Smermesnil 49.84 1.49 6.6 9.0 11.4 507.3 1315.1 2254.9 7877002.1 6215570.8
Le Buret 47.54 -0.29 6.1 7.9 9.8 374.6 886.0 1646.8 5772338.3 4884633.6
Puchot 49.41 1.59 6.2 8.3 10.4 397.6 1025.7 1909.9 6428158.2 5291480.0
Tillières 47.9 -1.8 6.3 8.3 10.5 426.6 1025.9 1964.1 6351356.6 5661760.3
Plaine
Conlinoise 48.7 0 5.5 7.5 9.9 261.8 759.6 1688.9 5216721.6 4322866.1
Merisier 47.25 -1.24 5.5 7.2 8.8 261.7 660.9 1206.6 4206595.1 3791574.5
Malaucourt 48.5 6.2 5.7 7.6 9.8 306.0 791.9 1648.7 5430143.2 4341619.7
Daméraucourt 49.41 1.55 6.4 8.6 10.8 452.1 1147.9 2071.6 7141511.3 5735551.7
Table (2). Data for all the locations

Conclusion
Assessing wind resources is a crucial step before developing and installing wind farms. However, investors and
project developers are often only concerned about the financial aspects of the project. To ensure profitability, it
is important to analyse the wind resource, along with turbine technology and efficiency. This analysis typically
includes investigating the normal probability projection. The figure below shows the steps to access the risk.

17
 When investing in wind projects, financiers need to understand both the sources and number of losses and
uncertainties. These concepts are distinct from each other, with losses being estimates of known decreases in
energy output. Factors that contribute to losses include wake, electrical, plant availability, turbine performance,
environmental, and curtailment. In developed markets and regions with well-known data, uncertainties can be
quantified using probability of exceedance levels such as P50, P75, P90, and P95. These parameters are crucial
for financial project evaluation and risk assessment, with investor decisions typically based on P50 and P90
levels.

To accurately assess the P50 value, better instrumentation, sensors, and long-term data assessment are needed.
The level of uncertainty directly impacts the difference between P50 and P90 levels of energy generation. The
smaller the uncertainty, the smaller the difference between P50 and P90, which in turn reduces financial risks
for banks and investors. This leads to more attractive financing conditions for wind projects.

References
[1] Malaucourt-sur-Seille (France) - Wind farms - Online access - The Wind Power

[2] Malaucourt Sur Seille wind farm - Global Energy Monitor (gem.wiki)

[3] Guidelines for Wind Resource Assessment: Best Practices for Countries Initiating Wind Development
(adb.org)

[4] Understanding the strategic imperatives of financial forecasting in your organization - 1CloudHub: Digital
Transformation – Advisory | Solutions | Implementation

[5] Introduction to the NetCDF4 Hierarchical Data Format | Earth Data Science - Earth Lab

[6] c++ - Bilinear interpolation artifacts - Stack Overflow

Appendix
Interpolation of data

18
 P50, P75 and P90 Plots

Time Series analysis plot

19