Uncertainties in the estimate of wind energy

production
LIRA, A. G.1; ROSAS, P.A.C 2; ARAÚJO, A. M3; CASTRO, N. J.4
1
Federal University of Rio de Janeiro – UFRJ; Economy Institute- IE; Study Group of Electric
Sector – GESEL; 250 Pasteur Avenue, Rio de Janeiro, Brazil; [email protected]
2
Federal University of Pernambuco – UFPE; Department of Electrical Power System
Engineering; Academico Helio Ramos Street– Cidade Universitária, Recife, Brazil;
[email protected]
3
Federal University of Pernambuco – UFPE; Department of Mechanical Engineering;
Arquitetura Avenue – Cidade Universitária; Recife; Brazil [email protected]
4
Federal University of Rio de Janeiro – UFRJ; Economy Institute- IE; Study Group of Electric
Sector – GESEL; 250 Pasteur Avenue, Rio de Janeiro, Brazil; [email protected]

ABSTRACT
The Annual Energy Production (AEP) estimated over the lifecycle of the project is one of the
most important factors to determine the profitability of wind power project. The methods used to
estimate the AEP in a wind farm requires an assessment of the uncertainties associated at all
steps. To finance a wind power project, banks requires that the developer submit the uncertainties
related to the estimation of AEP's wind farm, to mitigate errors and increase the project
reliability. The appropriate assessment of uncertainties is critical to determine the feasibility and
risk in developing a wind energy project. This study presents the main sources of uncertainty in
the energy estimate process in wind farms. This information is important for the correct analysis
of the economic viability of the project.

KEYWORDS: Wind Power, Uncertainty, Annual Energy Production

1 INTRODUCTION

The energy production in a wind farm follows a stochastic principle and, as such, requires a
statistical analysis in which production estimates should be associated with occurrences
probabilities.
An uncertainty analysis is often performed as part of a wind farm energy yield assessment. The
economic viability of a wind farm requires an analysis of the risks associated with the production
uncertainties.
The Section 3 this paper describes the main sources of the uncertainties in an energy assessment.
Each wind farm development uncertainties must be individually determined and then calculated
for the entire project. The are several methods, such as the IEC method (IEC 61400-12 Power
Performance Testing) for the evaluation of measurement uncertainty or the Monte Carlo method,
which lead to different results related to the different processes.
An interesting way to present the project uncertainties is by giving the probabilities of
exceedance in terms of expected annual production of the wind farm.
In the financing process of wind farms, banks have specific requirements in order to ensure that
the energy estimate has the smallest error margin possible. In Brazil, due to of auctions, the
Energy Research Company - EPE, demands a declaration issued by an independent certifying,
declaring the Physical Guarantee (GF), which is the annual energy availability for each wind
farm competing in the energy auction.
In order to mitigate the risk that energy production be less than the one on contract, the Physical
Guarantee of the wind power generated must be calculated taking into account all sources of
uncertainties in the project, so that the certified energy can bear a 90% probability, being attained
or exceeded. This value is called P90.
According to Tolmasquim et.al. (2013) the economic viability of wind energy production within
the regulatory framework of the Brazilian electricity market emerged for the need for a set of
specific rules, aimed at the following objectives:
• To imbue the business agent with the effective production of the energy contracted;
• To minimize cost of energy, reducing the financial cost of projects, mitigating
uncertainty in revenues from energy sales;
• Encourage the efficient procurement of the wind farm; and
• Reduce the risk of non-compliance of the contracted energy amount.
With the current energy auctions rules, entrepreneurs are penalized for producing below the
contracted amount of energy, pursuant to a tolerance margin.
Reducing uncertainty by increasing the quality of the criteria design is the only way to keep the
financial risk of a wind farm within acceptable limits for financiers, besides providing greater
security in meeting energy demand.
It is important to understand the main sources of uncertainties in a wind farm project in order to
reduce their magnitude and then accurately calculate its impact on yield forecasts.

2 LITERATURE REVIEW

There are many references to research about uncertainties in the estimation of the wind farm
energy production.
To name a few, in Lira (2012) is presents the main sources of the uncertainty in wind energy
production, in Pedersen et.al. (2006) is presents an analysis of the performance of some types of
anemometric, in Mortensen et.al. (2006), Corbett (2012) and Mortensen et.al. (1997) are
presents the uncertainty about the wind flow simulation models and in Lackner, M. A. (2008) is
present new approach for wind energy site assessment considering uncertainty.
3 OBJECTIVE

The calculation of the estimative energy production from a wind farm is subject to uncertainties
that must be accounted for in order to assess the risk of investments based on the accuracy of the
estimated energy production.
The main goal of this paper is to present the main sources of uncertainty in energy production
estimate process for wind farms in order to identify the expected improvement in energy
reliability and reduce the financial risks of the projects.

3 MAIN SOURCES OF UNCERTAINTIES

The main sources of uncertainties can be split into two groups: Wind Resource Uncertainty and
Energy Production Uncertainty.

3.1 - Wind Resource Uncertainty

This uncertainty has to do with limitations in the measurement process at meteorological tower.
Included the uncertainties associated with the type of the sensor, installation and calibration of
sensors, location of the towers, etc.

To turn the uncertainty of the wind resource into uncertainty in energy production the sensitivity
factor is required. The sensitivity factor corresponds to the variation in energy production caused
by wind variation, it is specific value for each project. Energy production and wind speed shows
no linear relationship.

3.1.1 - Sensor accuracy

The quality of results is directly dependent on the quality of equipment and the way they are
installed at the meteorological tower. The costs of a high-quality measuring system and its
appropriate installation are small when compared with the costs of a wind farm. It is
recommended the use of the anemometers models This First Class or Vector, which despite their
higher prices, they have reduced measurement uncertainties when other sensors on the market are
taken into account. The Fig. 1 shows the main models of the cup anemometers.

Fig. 1- Main models of the cup anemometers. Source: Pedersen et.al. (2006)
According to IEC 61400-12-1 (2005) the Eq.1 gives the operational standard uncertainty
𝑘 Eq.1
𝑢𝑖 = (0,05 𝑚⁄𝑠 + 0,005 ∗ 𝑈𝑖 ) ∗
√3
Where: 𝑈𝑖 is wind speed bin and k is classification number

According to Pedersen et.al. (2006) the simple uncertainty range in terms of wind speed and
associated with the instrument's accuracy for an isolated sensor comprises values between
approximately 1% and 6%.

3.1.2 - Sensor calibration

One important aspect concerning quality warranty of wind measurement is the anemometers
calibration through an appropriate wind tunnel. There are studies that show uncertainties greater
than 3.5% from anemometers calibrated in various wind tunnels. For this reason, MEASNET –
Measuring Network of Wind Energy Institutes – has issued a measurement method for measuring
cup anemometers calibration, especially custom-made for wind energy.
By following this practice, institutions provide guarantee that the wind tunnels used will not
differ more than 0,5% in the reference wind speeds and, thus, such a procedure will provide a
small and controlled uncertainty from the anemometers certified by the aforementioned method.
Currently, the great majority of research and wind power assessment institutions require that
anemometers have calibration certificates issued by institutions that have the MEASNET stamp,
that is, they observe the calibration standard set by that institution.
The use of individually calibrated anemometers poses a direct impact in reducing the wind speed
measurement uncertainty.
According to Coquilla et.al. (2008) the mean relative uncertainty on the calibration of various
cup and propeller anemometer models is present in the Table 1.

Table 1 - Mean relative uncertainty.

Mean Relative
Cup Anemometer Model
Uncertainty (%)
NRG #40 1,48
NRG IF3 1,66
Risoe Cup 1,43
R.M. Young Propeller 0,50
R.M. Young Wind Monitor 0,75
R.M. Young Wind Sentry 1,02
Second Wind C3 1,64
Thies First Class 2,04
Vaisala WAA252 1,98
Vector A100LK 2,06
Vestas Cup 1,09
Source: Coquilla et.al. (2008)
3.1.3 - The uncertainty due to assembly of the sensor

The anemometers and direction sensors (wind vane) should be fixed in the tower by means of
rigid booms, so there is no vibration in the sensors and thus the data measurement will suffer no
interference. The length of the boom-mounted must follow pursuant to IEA’s recommendations -
International Energy Association. The separation between the tower and the sensors should
reflect the level of uncertainty considered acceptable.
Fig. 2 shows iso-speed plot with flow disturbance because of the proximity to the tower. On the
left is a tubular tower and to the right a triangular lattice tower.

Tubular Tower Triangular Lattice Tower

Fig. 2 – Meteorological tower interference in the wind flow. Source: IEA (1999)
In order to minimize tower interference in the anemometer, this equipment must be apart from
the met tower by a minimum distance, and positioned where the wind speed isolines interference
reaches the closest value to the unit, making use of the prevailing direction of wind as reference.
The IEA´s standard guidelines recommends for tubular towers and a 0,5% error, the minimum
distance between the sensors and the tower be 8,5 times the diameter of the tower, measured
from the center of the tower. For lattice towers and a 0,5% error, the distance should be at least
5,7 times the width of the face. It is recommended, however, that the boom-mounted should not
be much bigger than this measure to reduce vibrations.

3.1.4 - Uncertainty in the long-term wind prediction

The wind displays a stochastic behavior where a significant interannual variability is observed,
that is, wind speed average may vary from year to year.
Wind measurements in short periods (1-3 years) are not indicative of long-term wind resource
due to interannual variability.
Therefore, to assesses correctly a local wind potential a long period of data is required to reduce
the error associated with these wind behavior changes over the years. Thus, to reduce errors in
the estimate of the wind farm energy production, a data correction measured on site with long-
term data is carried out. This correction improves the estimate of long-term wind speed, but also
brings uncertainty in the process.
To analyze the uncertainty in the long-term wind prediction is important take into account the
uncertainty on historical wind conditions and the uncertainty in future wind variability.
The uncertainty in the historical wind conditions is related with the correlation between the target
site (measured data) and the reference station (long-term data). The weaker the correlation with
the reference station, the larger the uncertainty in the adjusted long-term wind resource at the
target site, some estimates are given in the Table 2.

Table 2 - Wind Speed correlation uncertainty as a function of R2

Correlation coefficient (R2) Wind Speed correlation uncertainty
> 0.9 <1%
0.9 - 0.8 1 - 2%
0.7 - 0.6 3 - 5%
Source: GL Garrad Hassan (2011)

The uncertainty in future wind variability, in consideration of conducted studies NYSERDA

(2010), should be approximately 1,4% (10 years) and 2,2% (25 years).

3.1.5 - Uncertainty in the wind flow simulation

The wind flow model is not always able to describe the wind behavior of the meteorological
towers to the location of turbines. The terrain complexity, local roughness, the existence of
obstacles and the distance of each turbine from the meteorological towers are among the factors
that determine the magnitude of uncertainties. The range of uncertainty can be very wide, but a
typical range is 3% - 6%.

3.1.6 – Other

Other sources of uncertainties in wind resource must be taken into consideration: Uncertainty in
vertical wind extrapolation, uncertainty in the numerical simulation of wakes, uncertainty of
wind data availability, etc.

3.2 - Energy Assessment Uncertainty

3.2.1 - Uncertainty due to Power curve

The power curve of a wind turbine is the curve that indicates the power output for each specific
wind speed, and thus is one of the main parameters for estimating energy production. Due to
terrain characteristics, the wind flow often displays different features from those in which the
characteristic curve of the wind turbines had been designed. This may reflect different power
curve output. Variables such as turbulence and topography can play a significant role in the
variation of the power curve wind turbine.
When the power curve measurement test is carried out according to the international procedures,
the uncertainty typical is between 4 and 6%. If the power curve measurement test is not made,
the uncertainty of the power curve can be seen between 8% and 10%.

3.1.6 - Other

Uncertainty due to Electrical losses, Uncertainty due to Energy availability, etc.

4 UNCERTAINTY CALCULATION METHODS

There are two methods for calculating uncertainties: the deterministic method and the Monte
Carlo analysis.
According to Fontaine et.al. (2007) the deterministic method is based around the assumptions
that the different uncertainties are independent and that there is a linear relationship between the
input uncertainties and the output uncertainty. The various individual uncertainties are summed
using the Root Means Squared (RMS). This method does allow for the magnitude of the
individual uncertainties to be determined.
The Monte Carlo method for estimating energy uncertainties is a stochastic method simulating
the behavior of a physical system a large number of times. In a wind farm uncertainty analysis,
these simulations produce wind farm outputs while randomly varying the uncertainties according
to a defined probability distribution. The final uncertainty estimates are then determined from the
distribution of the simulated outputs. This allows for non-linear relationships between the
different uncertainties since the final uncertainty is not the result of summing the various
uncertainties.

5 ENERGY AND PROBABILITY OF EXCEEDANCE

The methodology used to obtain the total uncertainty of the project, the various sources of
uncertainty combined, may vary from company to company. Therefore, the same project, when
carried out by more than one company, may produce different overall uncertainty regardless the
use of the same data.
To properly estimate energy production, in addition to evaluate the project's uncertainties, it is
essential to consider all energy losses as the electrical loss, unavailability of wind turbine,
unavailability of electrical grid, wake loss, to name a few.
Following calculation of energy production and discounting all energy loss, the value of net
Annual Energy Production (AEP) is attained.
The net AEP and total uncertainty determine, respectively, the mean and standard deviation for a
normal Gaussian distribution. The absolute standard deviation is obtained by multiplying the
total uncertainty by net AEP.
The calculated net AEP is the value of the energy production called 𝑃50 , central energy
production estimate in the normal Gaussian distribution. This represents an energy value with a
50% probability of being exceeded.
In general, the probability of energy production distribution, assuming a normal Gaussian
distribution is given by Eq.2.
−(𝐸−𝐸𝑚 )2⁄
2𝜎2 Eq.2
1
𝑓(𝐸) = 𝑒
𝜎√2𝜋

Where:
f (E) is the probability of production being equal to the E energy [%];
𝐸𝑚 is the mean of normal Gaussian distribution; The net AEP with a 50% probability to
be exceeded; 𝑃50 ;
σ is the absolute standard deviation of the energy production estimate.
The Eq.2 is shown graphically in Fig. 3 indicating the 𝑃50 value.

Fig. 3 - Normal Distribution – Energy production probability

To use tabulated values, they must be converted into a normalized Gaussian distribution.
To know energy production with a specific probability level, normal distribution tables for
specific probabilities and the corresponding values of z need to be utilized.
With the 𝑃50 value, uncertainty total of the project and the z table probability, through of the Eq.3
it is possible to calculate the value in net energy production for the desired probability of
exceedance.

𝑃𝑥 = 𝑃50 ∗ (1 − 𝑧 ∗ 𝑈𝑛𝑐𝑒𝑟𝑡𝑎𝑖𝑛𝑡𝑦𝑇𝑜𝑡𝑎𝑙 ) Eq.3

Where:
𝑃𝑥 is the net energy production to desired probability of exceeded.
𝑈𝑛𝑐𝑒𝑟𝑡𝑎𝑖𝑛𝑡𝑦𝑇𝑜𝑡𝑎𝑙 is the total project uncertainty.
z is the value found in the probability table.

The z value is dependent on desired probability. The Table 3 shows z values for various
probability levels.
 Table 3 - Normal distribution table of specific probabilities and their corresponding z values
Probability of
exceedance z
(%)
99 2,326
95 1,645
90 1,282
85 1,036
84 1,000
80 0,842
75 0,674
50 0
25 0,674
10 1,282
1 2,326

It is important to notice that the total uncertainty is related to the energy value in 𝑃50 .
The net AEP in 𝑃90 translates a 90% probability of being attained or exceeded.
It is recommended that the total uncertainty of the project should be around 15%.
The higher the value of total uncertainty, the higher the difference between 𝑃50 and the other
levels of probability of exceedance.

5.1 – Examples of Probability of exceedance

Following are three examples with the same amount of energy in 𝑃50 , but with different values
for total uncertainty. The energy values in 𝑃75 and 𝑃90 (75 % and 90 % probability of
exceedance) are used to show the impact caused by overall uncertainty.

5.1.1 – Example 1: P50 of 120 GWh/year and total uncertainty of 10%

The Table 4 shows a project with energy in 𝑃50 equal to 120 GWh/year and 10% of total
uncertainty. In this example, the energy values in 𝑃75 and 𝑃90 are respectively 7% and 13% lower
than the value of energy in 𝑃50 .

Table 4 - Example 1: P50 of 120 GWh/year and total uncertainty of 10%

P50 P75 P90
Uncertainty
(GWh/year) (GWh/year) (GWh/year)
120 10% 112 105
Difference from P50 -7% -13%

The Fig. 4 shows various levels of probability of exceedance for this example.
 220
200 P50 = 120 GWh/y and Uncertainty = 10%

Net Energy (GWh/year)

180
160
140
120
100
80
60
40
20
0 10%
15%
20%
25%
30%
35%
40%
45%
50%
55%
60%
65%
70%
75%
80%
85%
90%
95%
99%
1%
5%

Probability of Exceedance

Fig. 4- Probability of exceedance: P50 of 120 GWh/year and Uncertainty of 10%

5.1.2 – Example 2: P50 of 120 GWh/year and total uncertainty of 15%

The Table 5 shows a project with energy in 𝑃50 equal to 120 GWh/year and 15% of total
uncertainty. In this example, the energy values in 𝑃75 and 𝑃90 are respectively 10% and 19%
lower than the value of energy in 𝑃50 .

Table 5 - Example 2: P50 of 120 GWh/year and total uncertainty of 15%

P50 P75 P90
Uncertainty
(GWh/year) (GWh/year) (GWh/year)
120 15% 108 97
Difference from P50 -10% -19%

The Fig.5 shows various levels of probability of exceedance for this example.

220
200 P50 = 120 GWh/y and Uncertainty = 15%
180
Net Energy (GWh/year)

160
140
120
100
80
60
40
20
0
10%

70%
15%
20%
25%
30%
35%
40%
45%
50%
55%
60%
65%

75%
80%
85%
90%
95%
99%
1%
5%

Probability of Exceedance
Fig. 5 - Probability of exceedance: P50 of 120 GWh/year and Uncertainty of 15%
5.1.3 – Example 3: P50 of 120 GWh/year and total uncertainty of 30%

The Table 6 shows a project with energy in 𝑃50 equal to 120 GWh/year and 30% of total
uncertainty. In this example, the energy values in 𝑃75 and 𝑃90 are respectively 20% and 38%
lower than the value of energy in 𝑃50 .

Table 6 - Example 3: P50 of 120 GWh/year and total uncertainty of 30%

P50 P75 P90
Uncertainty
(GWh/year) (GWh/year) (GWh/year)
120 30% 96 74
Difference from P50 -20% -38%

The Fig. 6 shows various levels of probability of exceedance for this example.

220
200 P50 = 120 GWh/y and Uncertainty = 30%
Net Energy (GWh/year)

180
160
140
120
100
80
60
40
20
0
10%
15%
20%
25%
30%
35%
40%
45%
50%
55%
60%
65%
70%
75%
80%
85%
90%
95%
99%
1%
5%

Probability of Exceedance

Fig. 6 - Probability of exceedance: P50 of 120 GWh/year and Uncertainty of 30%

6 CONCLUSION

It is important to properly quantify the uncertainties of a wind project because they may represent
significant variations in energy production. The uncertainty analysis is, therefore, paramount in
assessing economic viability of a wind power project.
The extra costs for accurate wind monitoring are relative very small compared to a high
investment in a wind energy project.
It is recommended to use first class anemometers and they need to be correctly calibrated.
Multiple measuring towers are very important to reduce the uncertainty. The maximum distance
between proposed turbine location and meteorological tower should be lower than 6km for flat
terrain and 2km for complex terrain.
The proper wind flow model is important to reduce the uncertainty. The linear model is
recommended to flat terrain and neutral climatic conditions. For complex terrain, usually CFD
model is recommended.
It is essential to define a standard methodology for the calculation of uncertainties in energy
production on wind farms in order to avoid significant differences in the calculated energy from
different independent certifiers.

REFERENCES
Asian Development Bank (2014). Guidelines for Wind Resource Assessment: Best Practices for
Countries Initiating Wind Development.
Coquilla, R. V. & Obermeier, J. Calibration Uncertainty Comparisons Between Various
Anemometers. In American Wind Energy Association AWEA, May 2008.
Corbett, J.F. CFD can consistently improve wind speed predictions and reduce uncertainty in
complex terrain. In EWEA 2012 Annual Event, 16-19 April 2012.
Fontaine, A. & Armstrong, P. (2007). Uncertainty Analysis in Energy Yield Assessment. In
EWEC Annual and Offshore Events, December 2007.
GL Garrad Hassan (ed. lit.). (2011). Uncertanty Analysis. In GL Garrad Hassan (ed. lit.),
WindFarmer Theory Manual (36-39). England: Garrad Hassan & Partners Ltd.
IEA, International Energy Agency. (1999). Recommended Practices for Wind Turbine Testing -
Part 11 Wind speed measurements and use of cup anemometry, Second print 2003
IEC 61400-12-1 (ed. lit). (2005) Annex C: Evaluation of uncertainty in measurement. In IEC
61400-12-1 (ed. lit), Wind turbines – Part 12-1: Power performance measurements of electricity
producing wind turbines, (39-40). Switzerland: IEC.
Lackner, M. A. (2008). The streamlined site assessment methodology: a new approach for wind
energy site assessment. Ph.D. Thesis, University of Massachusetts, Massachusetts, EUA.
Lira, A. G (2012). Analysis of Uncertainties in the estimate of the wind farms energy production.
Master's degree dissertation, Federal University of Pernambuco, Recife, Brazil.
Mortensen, N. G. & Petersen, E. L. Influence of topographical input data on the accuracy of wind
flow modeling in complex terrain. In European Wind Energy Conference & Exhibition, October
1997
Mortensen, N. G., Anthony J. B. & Ioannis A. (2006). Improving WAsP Predictions in (Too)
Complex Terrain.
NYSERDA (ed. lit.). (2010) WIND RESOURCE ASSESSMENT HANDBOOK. New York
State Energy Research and Development Authority.
Pedersen, T.F.; Dahlberg, J.-Å. & Busche, P. (2006). ACCUWIND- Classification of Five Cup
Anemometers According to IEC61400-12- 1. Denmark: Risø National Laboratory.
Tolmasquim, M. T. & Farias, J. C. M. (coords.) (2013) Studies for the expansion of generation.
Rio de Janeiro, RJ: EPE.
Tolmasquim, M. T. & Farias, J. C. M. (coords.) (2014) Studies for the expansion of generation.
Rio de Janeiro, RJ: EPE