Energy Conversion and Management 107 (2016) 22–33

Contents lists available at ScienceDirect

Energy Conversion and Management

journal homepage: www.elsevier.com/locate/enconman

Technical, economic and uncertainty modelling of a wind farm project

Svetlana Afanasyeva a,⇑, Jussi Saari a, Martin Kalkofen b, Jarmo Partanen a, Olli Pyrhönen a
a
LUT School of Energy Systems, Lappeenranta University of Technology, Lappeenranta 53850, Finland
b
Department of Mechanical Engineering, Technical University of Munich, Garching b. München 85748, Germany

a r t i c l e i n f o a b s t r a c t

Keywords: Wind is one of the fastest-growing renewable sources of electricity. As with most renewables, the eco-
Wind farm nomic viability relies on public subsidies. Even among renewables, the economic evaluation of wind pro-
Financial modelling jects is particularly challenging due to the unpredictability of the energy source, wind, and the high
Risk assessment
sensitivity of the project profitability to changes in a number of parameters. This makes the uncertainty
Monte Carlo simulation
analysis an important topic of research in the wind power sector.
This study presents a method of combining two uncertainty analysis methods, sensitivity study and the
Monte Carlo method, together with a technical and economic model of a wind farm, in an effort to
improve the understanding of the practical effects of the uncertainties, and how they affect the financial
risks of wind projects.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction is important to be aware of the scope of the analysis to use right

Among renewables wind power is one of the most popular
sources of energy. The worldwide wind market grew by more than
25% annually during the last decade and future growth is expected
to show a similar trend [1]. One of the main advantages of wind
power is zero fuel cost and thereby independence from fuel price
fluctuations and supply security concerns. However, at the same
time the volatile and unpredictable nature of the wind also creates
a considerable financial risk to the project and the economic envi-
ronment carries its own risks as well.
Before investing in a wind farm project, an investor will want to
be sure that the project is feasible and not unacceptably suscepti-
ble to risks. The aim of the paper is to create a technical and eco-
nomic model of a wind farm and to incorporate a risk analysis to
provide a tool for predicting the economic performance of a wind
farm within a chosen uncertainty level.
Extensive manual by National Renewable Energy Laboratory
(NREL) [2] gives a guideline for economic evaluation of the renew-
able energy technologies, providing a comparative analysis of the
main economic metrics used in the industry: simple payback per-
iod (SPB), discounted payback period (DPB), benefit-to-cost ratios
(B/C), net-present value (NPV), total life-cycle cost (TLCC), internal
rate of return (IRR), cost of energy (COE), levelized cost of energy
(LCOE), etc. Each of the metrics has advantages and drawbacks: it
⇑ Corresponding author. Tel.: +358 465928849.
E-mail address: Svetlana.Afanasyeva@lut.fi (S. Afanasyeva).
metric for right application.
Several authors have studied the economics of wind farm pro-
jects. Similarly to [2], work [3] also discusses the economic metrics
listed above and some others, focusing particularly on the wind
energy industry. Authors propose using several metrics for the eco-
nomic evaluation of the project, as they can give different results.
The predominant economic analysis techniques are COE [4–6],
LCOE [7] and NPV [8]. The LCOE and NPV use net cash flows dis-
counted to the present.
In this work the economics of the project are considered in
terms of NPV obtained from the evaluation of investment costs,
operation and maintenance costs and income from the sold elec-
tricity. NPV is considered as the most explicit and robust method
for project financial analysis [3]. The results are validated against
a period of plant data from an operating wind farm. The disadvan-
tage of the NPV method is that it is based on forecasts of future
costs, prices and interest rates that may not hold true for the life-
time of the project, and it does not consider uncertainties of these
assumptions.
Common methods to consider uncertainty in the standard dis-
counted cash flows (DCF) techniques are scenario and sensitivity
analysis. These methods give some indication of the risk; however,
they do not provide clear measure of project risk, as they do not
produce an estimate about probable net profit. The Monte Carlo
(MC) simulations enable presenting a better picture of the
expected outcome and the related uncertainties. The MC method
yields stochastic predictions, provided that information about the
probability distributions of the uncertain variables is known. Most

http://dx.doi.org/10.1016/j.enconman.2015.09.048
0196-8904/Ó 2015 Elsevier Ltd. All rights reserved.
 S. Afanasyeva et al. / Energy Conversion and Management 107 (2016) 22–33 23

Nomenclature

A surface area (m2) q density (kg/m3)

AEP annual energy (MW h) r standard deviation
C cost (k€)
c Weibull scale parameter (m/s) Subscripts and superscripts
cT thrust coefficient COE cost of energy
D diameter (m); direction (°) d direction
f probability density function eff effective
h height (m) hub at the hub of the turbine
I income (k€) in cut-in
k Weibull shape parameter inv investment
NPV net present value (k€) LCOE levelized cost of energy
P power (MW) O&M operation and maintenance
p interest rate op operation
p probability out cut-out
T time (h) overlap overlapping
V wind speed (m/s) rated rated by manufacturer
SI sensitivity indicator
Greek symbols std standard
a wake decay constant SV switching value
C Gamma function swept swept by the wind turbine blades
c wind shear t time
e error VaR value at risk
j discount factor x at position x

commonly the normal distribution is assumed, but a number of
important uncertain parameters are in fact known to be not nor-
mally distributed. In wind farm research field uncertainties have
been considered by some authors, using both sensitivity analysis
[7,9] and MC simulations [10,11].
The work by International Energy Agency [7] recommends a
procedure for the classification of wind projects by levelized pro-
duction cost (LPC), summarising the input parameters and their
uncertainties, and proposes simple methodology to estimate confi-
dence interval of LPC. The parameters are assumed to be indepen-
dent and have Gaussian distributions; the overall uncertainty is
found using the expected values and variances of each input
parameter.
According to report [8] wind resource, electricity price, future
variable costs and discount rate are the main factors contributing
to uncertainty in the profitability of the wind farm. The report
investigates the influence of uncertainty in wind resource on the
optimization of the wind farm layout. The presented model also
incorporates a function describing the level of risk the investor is
willing to accept, based on Utility theory. It uses different scenarios
and their probability in long-term wind resource to find expected
values of NPV and utility function. Naturally this model demon-
strated that for risk averse investor the optimized layout is less
sensitive to wind resource variability, but at the same time prof-
itability is smaller compare to riskier investor.
In Ref. [10] a probabilistic cost model OWECOP (Offshore Wind
Energy Costs and Potential) is implemented. For this model a wide
range of relevant parameters was identified and their probability
distributions were constructed based on measurements and
experts’ opinions. To name a few, average wind speed and vertical
wind shear exponent were modelled using normal distribution,
Weibull shape parameter was assumed to follow uniform distribu-
tion, and for wind farm availability and array efficiency PERT distri-
bution was used.
In this paper to handle the disadvantages of the NPV method,
both the sensitivity analysis and the MC method are included to
account for uncertainties. MC simulations rely on the estimation
of the uncertain parameters from historical data; this may not
always be the most reliable way for estimating the future, how-
ever. For instance, electricity price trend may change due to
changes in policies and market situation, and due to climate
change wind resource at site can also change. It was therefore con-
sidered appropriate to accompany the MC method also with a sen-
sitivity analysis of uncertain variables, to provide additional
information about project economics under different conditions
for better assessment of the project.
In this work the MC simulations are performed using input
parameters whose distributions are known; wind resource param-
eters and power production of the wind turbine. For parameters
that are uncertain, but distributions and degree of uncertainty
itself is unknown, sensitivity analysis is applied.
2. Modelling
The framework structure used in this paper is shown in Fig. 1.
To evaluate the performance of a wind farm, both the economic
and technical aspects must be modelled. This is done separately
within the framework. Data of wind turbines and economics are
processed according to the wind farm design. The annual energy
production (AEP) is determined using the wind speed distribution
at the turbine locations, the power curve of the turbines and the
operational time per year. Together with economic data, the AEP
is the basis to calculate financial parameters in the economic
model. The Monte Carlo sampling module takes uncertainties into
account and is further explained in Section 3.
2.1. Economic model
The economic evaluation of the wind farm project is based on
the NPV, which represents the sum of cash flows during the life
time of the project discounted to the present. It is chosen as it gives
a clear picture of the profitability of the investment, and is also one
of the most commonly used economic evaluation methods. The
NPV is calculated from the investment cost C inv t operation and
maintenance (O&M) cost C O&M t and income It:
24 S. Afanasyeva et al. / Energy Conversion and Management 107 (2016) 22–33

Fig. 1. Structure of the underlying framework to simulate the wind park performance.

X
T the wind turbine power curve and the wind velocity frequency dis-
NPV ¼ ðIt  C inv t  C O&M t Þ  jt ; ð1Þ tribution at the turbine hub height, considering also the wind
t¼0
direction.
where jt is the discount factor at year t to discount the future val- For each wind direction d and each 1 m/s wind speed interval q
ues to the present, calculated from the discount rate p: from cut-in wind speed of turbine Vin to the cut-out speed Vout the
product of wind turbine power output Pq,i,t and probability of the
jt ¼ ð1 þ p=100Þt : ð2Þ wind speed V fd,q,i,t(V) and direction D pd,i,t are calculated. These
The investment costs consist of the cost of the wind turbines, are all summed up and multiplied by the annual operation time
electrical infrastructure such as cables and substation equipment, T op i;t to obtain AEP of year t of a single wind turbine i. Wind tur-
roads, planning and design cost, permits, rent of land, and civil bines affect each other’s power production as they slow down
works [12]. The investment costs of the turbines are estimated the wind speed behind the rotor, consequently reducing the power
according to [13]. It is assumed that all investments are made in production of the turbine(s) behind. This is accounted for with the
the first year of the project. wake effect wd,q,i,t. The AEP is obtained from Eq. (3),
Operation and maintenance includes costs of necessary repairs
X
D X
V out
and replacements, as well as supervision and administration costs. AEPi;t ¼ T op i;t pd;i;t  f d;q;i;t ðVÞ  P q;i;t ðVÞ  wd;q;i;t : ð3Þ
Estimating the O&M cost of a wind farm is challenging, as historical d q¼V in
data from life time operation is still lacking. Wind farm owners
often make fixed price service contracts per MW hel produced with In this study a wind rose of 12 sectors is used to model the wind
independent contractors over the project life time, allowing the conditions. The wind rose contains statistical information of the
owner to hedge against performance risks. In this paper O&M cost wind speed in each sector d. The Weibull probability density func-
is assumed to be a variable cost estimated at 12 €/MW hel [12]. tion [16],
The NPV is very sensitive to the assumed interest rate and the  k1
k V
eð c Þ ;
V k
price of electricity, which have a high degree of uncertainty. In f ðVÞ ¼ ð4Þ
many European countries wind power sector is supported by dif- c c
ferent incentives, such as feed-in tariff, feed-in premium, and green is used as a model of the wind speed distribution in each sector. The
certificates [14] that mitigate investment risk. parameters k and c of Eq. (4) depend on the mean wind speed V and
the standard deviation rV [16].
2.2. Technical model The wind speed V at the hub height h, needed to calculate the
power production, is obtained from the power law [17]
The annual energy production (AEP) of the wind farm is deter- c
mined similarly to IEC standard 61400-12-1 [15] as a function of V=V ref ¼ ðh=href Þ ; ð5Þ
 S. Afanasyeva et al. / Energy Conversion and Management 107 (2016) 22–33 25

where c is the wind shear and subscript ‘ref’ refers to the reference
height at which the wind speed Vref is known.
The frequency of wind f(V) in Eq. (3) and parameter k in Eq. (4)
are obtained from the Finnish Wind Atlas [18]. The Finnish Wind
Atlas divides the country to a 2.5  2.5 km2 grid and contains infor-
mation on wind direction frequencies, mean wind speed, and Wei-
bull parameters k and c for heights from 50 m to 400 m for each
grid square. Up to the height of 150 m data are given at 25 m inter-
vals. Values of f and k are obtained by interpolation. The Weibull
parameter c, however, is not taken from the Wind Atlas, but
obtained from [16]

V
c¼ ; ð6Þ
Cð1 þ 1=kÞ
where C the gamma function.
The wind speed downstream of each turbine is modelled using
a modified version of the Jensen wake model [19]. The volume of
the reduced wind speed forms a truncated cone opening behind
each turbine. The radius of the cone increases by factor a with Fig. 3. Thrust coefficient over wind speed.
the distance x, as shown in Fig. 2 below.
In this paper a commonly used value of a = 0.075 [20] is used,
translating to a cone opening angle of 9°. At a distance x behind To calculate the wake effect of multiple turbines, i.e. the effect
on a turbine inside several wake cones, the resulting speed is
a turbine i the cone diameter Dx i and the wind speed inside Vx i
obtained by ‘Energy balance’ approach [20]
are obtained from
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ! XNi  2 
1 1  cT i Vi ¼ V 20  V j  V 2i;j ; ð10Þ
Vx i ¼ V0 1  ; ð7Þ j¼1
ð1 þ 2ax=D0 i Þ2
where Ni is the number of wind turbines influencing the turbine i;
Dx i ¼ D0 i ð1 þ 2ax=D0 i Þ; ð8Þ V0 the free undisturbed speed, Vj the speed at the inlet of turbine
j; Vi,j is the wind speed at turbine i due to the wake of the turbine j.
where D0 i is the turbine i diameter, cT i the thrust coefficient and V0 Power produced by a turbine is determined from the manufac-
the free-stream wind speed. turer’s power curve based on standard conditions (subscript ‘std’)
The thrust coefficient is the ratio of the thrust force acting on at air density 1.225 kg/m3, normalized to the site conditions with
the turbine and the product of the air flow dynamic pressure and the equation [15]
the turbine swept area. The coefficient depends on the wind speed. qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
In this paper the coefficient is modelled as 7/Vhub (Fig. 3) [21]. V site ¼ V std 3
qsite =qstd : ð11Þ
If a turbine is only partly in the wake of another turbine located
upstream, the wake loss is reduced according to the percentage of
the turbine frontal area, which is inside of the wake cone. Thus the 3. Uncertainties
wake loss is multiplied by the ratio of the overlapping area
between the circles of wake cone and rotor and the complete Evaluating the economic and technical performance of a wind
swept area of the downstream turbine [22]: farm project involves several uncertainties arising from a number
V x i Aoverlap of sources. The natural variability of wind speed, direction and
V eff i ¼ : ð9Þ density introduces a significant unavoidable uncertainty to any
V 0 Aswept
evaluation. The wind uncertainty can be divided in four categories
This leads to a smooth transition between a turbine inside and [23]: (1) measurement, (2) long-term resource estimation, (3) site
outside the wake cone. assessment, and (4) wind resource variability.

Fig. 2. Jensen model for the wind speed behind a turbine.

26 S. Afanasyeva et al. / Energy Conversion and Management 107 (2016) 22–33

Model uncertainties are inevitable, as any model is only an

approximation of the physical reality. Simplifications made in the
model include, for example, the discretization of the wind rose to
12 directions, the interpolation of the wind speed from the avail-
able data points to the turbine location and its hub height, and
the estimation of mechanical and electrical losses between the
power extracted by the turbine blades and the electrical power
transmitted to the grid.
The power curve itself also contains significant uncertainty.
Firstly, some information is inevitably lost as noisy data is com-
pressed to a single curve, and, secondly, the power curve is used
with hub height wind speed, although wind speed in reality varies
from the rotor disc bottom to top [23,24].
Finally, the profit depends on the availability and economic
environment, which again contains significant uncertainties. Esti-
mating availability is difficult, since publically available historical
data on wind farms is sparse and, due to the fast technological
Fig. 4. Standard deviation of the wind direction.
development of wind turbines, is often already obsolete when
becoming available. Uncertainties related to the economic envi- X
N X
D X
V out

ronment are typically significant and difficult to estimate in AEPt ¼ T op i;t ðpd;i;t þ ed Þ  f d;q;i;t ðVÞ  ðPq;i;t ðVÞ þ eP q Þ
the energy sector in general, but particularly challenging with i¼1 d q¼V in

wind due to the reliance on subsidies depending on political  wd;q;i;t ; ð12Þ

decisions.
Two common methods to account for uncertainties are (1) eval- where f is determined from Eq. (4) and by substituting Eq. (6) for c:
uating the sensitivity of the output parameters to input variables  ðkþek Þ
!ðkþek Þ1 V Cð1þ1=ðkþek ÞÞ
by sensitivity analysis and (2) calculating the probability distribu- ðk þ ek ÞCð1 þ 1=ðk þ ek ÞÞ V Cð1 þ 1=ðk þ ek ÞÞ 
ðVþe Þ
f ðVÞ ¼ e V ;
tion of the outputs, by using numerous random inputs according to ðV þ eV Þ ðV þ eV Þ
expected probability distributions, by means of MC simulation ð13Þ
[25].
The sensitivity analysis is fairly simple to implement, but does eV  ln Nð0; r2V Þ; ð14Þ
not give a clear picture of the likely output. The MC method is often
considered the best method for analysing uncertainties as it gives
more information about the expected output and its variability, if
ek  Nð0; r2k Þ; ð15Þ
the constraints and probability density functions (pdf) of the input
parameters are known. However, the reliance on the availability of ed  Nð0; r2d Þ; ð16Þ
the pdfs and sensitivity to their accuracy makes the method also
problematic. In order to obtain reliable results a very large number eP q  Nð0; r2P q Þ; ð17Þ
of calculations is needed, which may become excessively time-
with the errors e, the Gamma function C, the Weibull shape param-
consuming with computationally complex models.
In this paper the MC method is used with those parameters eter k, the mean wind speed V, the probability of direction pd and
whose pdfs are considered known to a reasonable degree of cer- efficiency due to wake wd,q,i,t. The subscript ‘i’ indicates the number
tainty. With the parameters that are known to be uncertain, but of the turbine and ‘t’ indicates the year. N is the overall number of
whose distribution and degree of uncertainty is uncertain, a sensi- turbines.
tivity analysis is applied. The MC method is used to calculate uncertainty resulting from
Investors want to be sure that the project is secure to be prof- the natural variability of the wind speed and power output, while
itable. Standard practice for big investment projects such as a wind the effects of uncertainties from for example measurement, scaling
farm is to follow not only the expected value (P50), but also look at or interpolation errors were evaluated by means of sensitivity
the value at risk (VaR) [26]. VaR is a commonly used risk measure, analysis.
showing that output will exceed this value with a given probabil- In the sensitivity analysis parameters are changed by 5% and
ity. For example, a 90% VaR of NPV = 0 means that the project will sensitivity indicators (SI) and switching values (SV) are calculated.
be profitable with 90% confidence. By applying sensitivity analysis The SI is the relative output change resulting from the input
to VaR it is possible to obtain more information about the risks of a change, and SV is the percentage of change of the input parameter
project. required to reduce the output to zero.
In this study the uncertainties of wind characteristic (speed and Sensitivity analysis for more realistic parameters ranges, based
direction) and power production of the wind turbine are calculated on actual 2013 cost and performance data [30], is performed. For
using the MC method. electricity price range is chosen based on [31].
It has been shown that the annual mean wind speed follows a
lognormal distribution [27,28]. The Weibull shape parameter fol- 4. Studied case
lows a normal distribution. The reported data from Horns Rev wind
farm [29] was analysed and by applying linear fitting it was found The wind farm analysed in this paper is situated in Finland. The
that direction has standard deviation of 0.005pd (Fig. 4). layout consists of seven 3 MW turbines, each having a turbine disc
The power curve standard deviation rP q depends on the wind diameter of 109.8 m and a hub height of 92.5 m. The site map is
speed as shown on Fig. 5 [24]. The uncertainty is assumed to follow depicted in Fig. 6, showing the location of the turbines (yellow1
a normal distribution.
To obtain the AEP needed for NPV calculation in the MC simu- 1
For interpretation of colour in Figs. 6, 11 and 12, the reader is referred to the web
lations, Eq. (3) is transformed to the following stochastic form: version of this article.
 S. Afanasyeva et al. / Energy Conversion and Management 107 (2016) 22–33 27

Fig. 5. Normalized uncertainty in the power curve to normalized wind speed at hub height [24].

Fig. 6. Locations of the turbines in the wind farm.

colour) and the closest grid point, where wind data is available (red
colour). The wind speed data (Fig. 7) is retrieved from the Finnish
Wind Atlas [18].
In Finland wind energy sector is supported by feed-in tariff of
83.5 €/MW h up to 12 years of operation [32].
The following economic data are used:
- Investment costs: 1500 €/kW of installed capacity [12,13].
- O&M cost: 12 €/MW h [12].
- Electricity selling price:
1–12 yrs feed-in-tariff of 83.5 €/MW h [32].
13–25 yrs long term expectation price at the market of 50 €/
MW h [31,33].
The life time of the project is presumed as 25 years, interest rate
is 5% and the availability is 90% throughout the year.
5. Results
The validation of the power production model is demonstrated
in this section. A sensitivity analysis of the expected values of a
deterministic and stochastic model is performed, but also the value
28 S. Afanasyeva et al. / Energy Conversion and Management 107 (2016) 22–33

Fig. 7. Wind data from Finnish Wind Atlas for the site [18].

Fig. 8. Wind rose according to measurements at the site in November 2013.

at risk (VaR) for 90% of certainty is obtained with the stochastic
model, as this value would likely be of the investors’ interest.
5.1. Validation of the technical wind farm model
The technical wind farm model is validated using one month
data of wind conditions and power production data of the wind
farm. The wind rose is shown in Fig. 8; the average wind speed
is 7.5 m/s. The power curve is adjusted to the air density of
1.27 kg/m3 typical for this area.
The power production model gives the result of 4111 MW h for
wind conditions measured in November 2013 which is 3.5% less
than the actual production. According to model calculations it
results in NPV being underestimated by 15%. Due to the short per-
iod of validation data available, it is not possible to be certain
whether this is a random or systematic error, but overall the result
can be considered to be in good agreement with the actual power
 S. Afanasyeva et al. / Energy Conversion and Management 107 (2016) 22–33
Fig. 9. Measured and modelled power production of the wind farm.
output of the wind farm (Fig. 9). Measured power output depicted
in grey bars, and modelled power output is in red bars.
A MC simulation is run with 100,000 samples; the results are
shown in Fig. 10. The results of repeated runs differ by less than
1%. The terrain in the studied case is complex with lakes, forest
and fields. For this type of the terrain the uncertainty values for
mean wind speed and Weibull shape parameter are taken 10.5%
and 11.6% respectively [23]. Uncertainties of the wind direction
and power curve are as described in Section 3 Fig. 5.
The expected value of AEP and NPV calculated with determinis-
tic and stochastic equations are shown in Table 1 below. It can be
seen from the results, that the expected values from stochastic and
deterministic models differ from each other. The VaR of NPV is
Fig. 10. Results of AEP and NPV Monte Carlo simulation.
29
significantly less than the expected value, and from Fig. 10 it is evi-
dent that it is also highly sensitive to uncertainties in parameters
and doesn’t follow normal distribution.
5.2. Sensitivity analysis of the deterministic model
The results of the sensitivity analysis are presented in Table 2
and Figs. 11 and 12. Red and grey colours indicate negative and
positive change of input parameters, respectively.
For the AEP the greatest sensitivity lies with the windspeed,
which is explained by its cubic influence on the basic law for gen-
erated power, P  V3. Because the power generated is limited to the
rated power for high wind speeds, the sensitivity stays below the
theoretical value of 16% from a 5% change (1.053 = 1.1576). When
the power curve is linearly altered throughout the whole wind-
speed range, the power production changes proportionally. The
SI is therefore 1. The effect of the air density on the power curve
is estimated from Eq. (11) and scales the power curve to lower
wind speeds, which occur more often. The Weibull shape parame-
ter of power density function of the wind speed shows a negative
sensitivity. When changing the shape parameter, the scale param-
eter in the model is adjusted simultaneously to maintain the same
mean wind speed. A lower shape parameter leads to a wider
spread. The overall power production benefits more from the
increased occurrence of higher wind speeds than it loses through
the also higher occurrence of lower wind speeds. The Weibull scale
parameter translates proportionally to the average wind speed and
is therefore not listed separately. The wake decay constant deter-
mines how much the power production of a wind turbine is influ-
enced in the wake of a turbine situated upstream and plays only a
minor role. The parameters regarding the finances and the project
such as installation costs, O&M costs, discount factor, electricity
30 S. Afanasyeva et al. / Energy Conversion and Management 107 (2016) 22–33

Table 1
Results of deterministic vs. stochastic model.

Expected AEP (MW h/a) Std AEP (MW h/a) VaR AEP (MW h/a) Expected NPV (k€) Std NPV (k€) VaR NPV (k€)
Deterministic 46789.1 – – 7451.4 – –
Stochastic 46965.9 4738.5 40944.3 7598.6 3944.8 2585.6
Difference 0.4% – – 2% – –

Table 3
Switching values with stochastic analysis.
Table 2 Parameter Expected AEP VaR AEP Expected NPV VaR NPV
Switching values with deterministic analysis.
Avg. wind speed 42.4 38.4 8.2 2.9
Parameter Deterministic AEP Deterministic NPV Weibull shape par. 466.0 316.1 90.6 24.0
Avg. wind speed 39.9 7.6 Air density 126.8 116.6 24.6 8.8
Weibull shape parameter 359.2 68.7 Wake effect 3.4  103 5.1  103 667.5 393.7
Air density 118.8 22.7 El. price 6.5  105 1.2  105 62.1 24.1
Wake effect 3.33103 637.5 Power curve 99.9 99.6 19.4 7.6
El. price  60.9 Cost 2.1  105 1.1  105 24.0 8.1
Power curve 100 19.1 O&M cost 1.5  105 1.7  105 98.8 38.4
Cost – 23.7 Life time 8.2  104 1.9  105 61.3 23.4
O&M cost – 94.2 Interest rate 3.4  105 1.7  105 46.2 18.2
Life time – 58.2
Interest rate – 45.2

relatively small difference of two large values, i.e. income and

price and project life time, do not have an influence on the actual
electricity production and have the sensitivity (indicator) zero on
the annual energy production.
The NPV is generally much more sensitive than AEP to the
change of parameters. For the NPV all cash flows are discounted
with time to their present value and summed up. The NPV is the
expenditures. Its relative sensitivity is thus much higher than that
of the AEP. Again a significant impact can be seen from the average
wind speed, with an SI of 13 leading to a SV of 7.6%, i.e. decreas-
ing the average wind speed by 7.6% will reduce the NPV to zero.
NPV is strongly influenced by such parameters as power curve
and air density with the SI of 5.2 and 4.4 respectively. A similar
influence is noted for the cost of installation. Among the other

Fig. 11. Sensitivity analysis of deterministic AEP with ±5% change of parameters.

Fig. 12. Sensitivity analysis of deterministic NPV with ±5% change of parameters.
 S. Afanasyeva et al. / Energy Conversion and Management 107 (2016) 22–33 31

Fig. 13. Sensitivity analysis of VaR AEP with ±5% change of parameters.

Fig. 14. Sensitivity analysis of VaR NPV with ±5% change of parameters.

Fig. 15. Sensitivity analysis of VaR NPV with more realistic ranges of parameters.

economic parameters also electricity price, lifetime and the dis-
count factor all show an absolute SI greater than 1. In this analysis
electricity price variation is not applied to first 12 years of opera-
tion, as fixed-price feed-in tariff is assumed. When variation is
applied from the first year of operation, the electricity price
strongly influences the NPV with an SI of 5.4 and it is the second
most influencing parameter after wind speed.
5.3. Sensitivity analysis of the stochastic model
The 90% VaR was analysed in the stochastic model, as this value
is often of investors’ interest. The result of this sensitivity analysis
for the expected value is similar to results for deterministic model.
As expected, the VaR behaviour is similar to the deterministic
model when input parameters change. Due to the noise of the
stochastic function used in the MC simulation AEP wrongly shows
small sensitivities to changes in economic parameters. Results of
the sensitivity analysis are shown in Table 3, Figs. 13 and 14.
Results show that even small changes in inputs reduce VaR NPV
to zero. NPV is especially sensitive to changes in average wind
speed value (SV = 2.9%), followed by power curve (SV = 7.6%)
and air density (SV = 8.8%). The correct estimation of the wind
resource at a site is therefore crucial. The wind farm project is also
sensitive to installation costs (SV = 8.1%). The wake decay constant
doesn’t show significant influence on the NPV and AEP. This, how-
ever, does not mean that the wake effect itself would be insignifi-
32 S. Afanasyeva et al. / Energy Conversion and Management 107 (2016) 22–33

Fig. 16. Extended sensitivity analysis of VaR NPV with more realistic ranges of parameters.

Table 4
Summary of uncertain parameters, their impact and uncertainty, rated from low to high.

Impact Uncertainty Comments

Average wind speed High Medium Sensitive to measurement errors, estimation, and possibly changing climate
Air density Medium Low Unlikely to change as long-term average
Wake effect Low low Depends on site complexity
Electricity price Low/medium High Depends on policy decisions, which are unpredictable; low if 12-year certain
feed-in tariff period without uncertainty, otherwise medium
Power curve Medium Medium Depends on wind speed, site complexity, air density
Cost Medium Low On-shore installation is typically quick and uncomplicated
O&M cost Low High Long-term experiences are still lacking
Life time Low Low Long-term experiences are still lacking
Interest rate Low Medium Depends on economic situation

cant for wind farms installations. For electricity price the SV is
24.1%, increasing to 8.7% when variation is applied from first
year.
Sensitivity analysis for stochastic model was extended to more
realistic ranges of some input parameters. Ranges for investment
and O&M cost, interest rate are based on [30]. Range for electricity
price is based on [31,33]. Interval values for average wind speed,
power curve are similar to [7] (see Figs. 15 and 16).
Results show that the investment and O&M costs, interest rate
start to play most important role in the decision making of the pro-
ject when uncertainty in this economic parameters is high. Simi-
larly electricity price start to play much more crucial role.
6. Conclusions
the 90% VaR of NPV is 34% of the expected NPV, and relatively
small changes in input parameters can reduce it to zero. The risk
of the project becoming unprofitable is thus clearly non-
negligible, and effort should be made to quantify and minimize
uncertainties.
The uncertain parameters considered in the study together with
an evaluation of their impact on performance and uncertainty are
summarized in Table 4. The impacts of the listed uncertain param-
eters are derived from the tornado graph of the VaR of NPV
(Fig. 14). The parameters can be classified into three categories
according to the sensitivities of the VaR of NPV to them:
(1) High for impacts above 100%.
(2) Medium for impacts between 100% and 30%.
(3) Low for impacts below 30%.

A quantitative risk assessment is carried out to understand the
impact of uncertainties on the financial performance. A MC simu-
lation is used to assess risks whose probability distributions are
known, such as wind and power production. Sensitivity analyses
are performed for parameters where the probability distributions
and variability range could not be known, such as future electricity
prices and interest rates. The results are validated against a period
of plant data from an operating wind farm and found to closely
match the measured plant data.
The key factors to the wind farm performance and economic
feasibility were shown to be wind speed, electricity price, installa-
tion cost and power curve. The performance is also sensitive to air
density, which is, however, unlikely to vary significantly. Particu-
larly the NPV is sensitive to changes; the studied case showed that
Uncertainties are also divided into three groups. Average wind
speed, price of electricity and power curve are the most important
parameters as they have a high combined impact on the project
and uncertainty.
References
[1] GWEC – global wind report. Annual market update; 2014.
[2] Short W, Packey DJ, Holt T. A manual for the economic evaluation of energy
efficiency and renewable energy technologies. NREL Tech Rep 1995:1–120.
http://dx.doi.org/10.2172/35391.
[3] Oliveira WS, Fernandes AJ, Gouveia JJB. Economic metrics for wind energy
projects. Int J Energy Environ 2011;3:1013–38.
[4] Zhang J, Chowdhury S, Messac A, Castillo L. A response surface-based cost
model for wind farm design. Energy Policy 2012;42:538–50. http://dx.doi.org/
10.1016/j.enpol.2011.12.021.
 S. Afanasyeva et al. / Energy Conversion and Management 107 (2016) 22–33
[5] Schallenberg-Rodriguez J. A methodological review to estimate techno-
economical wind energy production. Renew Sustain Energy Rev
2013;21:272–87. http://dx.doi.org/10.1016/j.rser.2012.12.032.
[6] Roy S. Economic assessment of the engineering basis for wind power:
perspective of a vertically integrated utility. Energy 2009;34:1885–97.
http://dx.doi.org/10.1016/j.energy.2009.07.044.
[7] Tande JO, Hunter R. Recommended practices for wind turbine testing and
evaluation. 2. Estimation of cost of energy from wind energy conversion
systems. 2nd ed. IEA; 1994.
[8] González JS, Rodríguez ÁGG, Mora JC, Burgos Payán M, Santos JR. Overall
design optimization of wind farms. Renew Energy 2011;36:1973–82. http://
dx.doi.org/10.1016/j.renene.2010.10.034.
[9] Frandsen S, Christensen CJ. Accuracy of estimation of energy production from
wind power plants. Wind Eng 1992;16:257–68.
[10] Herman S. Probabilistic cost model for analysis of offshore wind energy costs
and potential. ECN Tech Rep 2002:1–47.
[11] Salles ACN, Melo ACG, Legey LFL. Risk analysis methodologies for financial
evaluation of wind energy power generation projects in the Brazilian system.
Int conf probabilistic meth appl power syst 2004;2004. http://dx.doi.org/
10.1109/PMAPS.2004.242157.
[12] Krohn S, Morthorst P-E, Awerbuch S. The European wind energy association.
Econ Wind Energy 2005;3. http://dx.doi.org/10.1504/IJETP.2005.006769.
[13] Fingersh LJ, Hand MM, Laxson AS. Wind turbine design cost and scaling model.
NREL Tech Rep 2006:1–38.
[14] CEER status review on renewable and energy efficiency support schemes.
Brussels; 2015.
[15] IEC 61400-12-1:2005 power performance measurements of electricity
producing wind turbines. vol. 2005; 2005.
[16] Manwell JF, McGowan JG, Rogers AL. Wind energy explained: theory, design
and application. Wiley; 2010. p. 1–704.
[17] Wizelius T. Developing wind power projects: theory and practice. Earthscan
2007:1–290.
[18] Tammelin B. Finnish wind atlas. Winterwind; 2008. <http://www.tuuliatlas.fi/
en/> [accessed September 1, 2015].
33
[19] Katic I, Højstrup J, Jensen NO. A simple model for cluster efficiency. Eur Wind
Energy Conf Exhib 1986;1986:407–10.
[20] Thørgersen ML, Sørensen T, Nielsen P. WindPRO/PARK: introduction to wind
turbine wake modelling and wake generated turbulence. EMD Int A/S 2008.
[21] Frohboese P, Schmuck C, Hassan GLG. Thrust coefficients used for estimation
of wake effects for fatigue load calculation. Eur Wind Energy Conf 2010;2010.
[22] Rathmann O, Barthelmie R, Frandsen S. Turbine wake model for wind resource
software. Proc EWEC 2006;2006:1–12.
[23] Lackner M, Rogers A, Manwell J. Wind energy site assessment and uncertainty.
ASME Conf 2007;2007.
[24] Gallagher R, Elmore AC. Monte Carlo simulations of wind speed data. Wind
Eng 2010;33:661–74. http://dx.doi.org/10.1260/0309-524X.33.6.661.
[25] ADB. Handbook for the economic analysis of water supply projects. Asian
Development Bank; 1999.
[26] Paramod J. Wind energy engineering. McGraw-Hill; 2011. p. 1–330.
[27] Kalnay A. The NCEP/NCAR 40-year reanalysis project. Bull Am Meteorol Soc
1996;77:437–70.
[28] Afanasyeva S, Saari J, Kukkonen S, Partanen J, Pyrhönen O. Optimization of
wind farm design taking into account uncertainty in input parameters. Eur
Wind Energy Conf Exhib 2013;2013:1–10.
[29] Serrano González J, Burgos Payán M, Riquelme Santos J. Optimum design of
transmissions systems for offshore wind farms including decision making
under risk. Renew Energy 2013;59:115–27. http://dx.doi.org/10.1016/j.
renene.2013.03.024.
[30] Moné C, Smith A, Maples B, Hand M. Cost of wind energy. NREL Tech Rep
2013;2015:1–95.
[31] Fleten S, Maribu K, Wangensteen I. Optimal investment strategies in
decentralized renewable power generation under uncertainty. Energy 2007.
[32] Feed-in tariff – Energiavirasto n.d. <https://www.energiavirasto.fi/web/
energy-authority/feed-in-tariff> [accessed September 1, 2015].
[33] Samis M. An extended evaluation framework for nuclear power plants – a
dynamic DCF and real options approach. WPNE workshop 2013.