Professional Documents
Culture Documents
1 s2.0 S0306261915006030 Main
1 s2.0 S0306261915006030 Main
1 s2.0 S0306261915006030 Main
Applied Energy
journal homepage: www.elsevier.com/locate/apenergy
h i g h l i g h t s
a r t i c l e i n f o a b s t r a c t
Article history: We propose the computation of different wind energy production indicators and financial profitability of
Received 3 February 2015 potential wind power sites. The computation is performed by modeling the wind speed process as an
Received in revised form 29 April 2015 indexed semi-Markov chain to predict and simulate the wind speed dynamics.
Accepted 30 April 2015
We demonstrate that the indexed semi-Markov chain approach enables reproducing the indicators cal-
Available online 19 May 2015
culated on real data. Two different time horizons of 15 and 30 years are analyzed. In the first case we con-
sider the government incentives on the energy price now present in Italy, while in the second case the
Keywords:
incentives have not been taken into account.
Wind speed
Indexed semi-Markov chains
Ó 2015 Elsevier Ltd. All rights reserved.
Monte Carlo simulation
Financial indicators
http://dx.doi.org/10.1016/j.apenergy.2015.04.124
0306-2619/Ó 2015 Elsevier Ltd. All rights reserved.
G. D’Amico et al. / Applied Energy 154 (2015) 290–297 291
the basis of this approach is the evidence that past values of the producing wind energy. The computed financial indicators allow
wind speed keep informations about present value and future val- managing such risks.
ues. Many scholars have proposed new models that can allow the The paper is presented as follow: First, we show our research
prediction of wind speed minutes, hours or days ahead. Many of methodology, the database used and the technical details of the
these models are based on neural networks [16], autoregressive commercial wind turbine. Second, the ISMC model is shortly
models [17,18], Markov chains [19–24], hybrid models where the described as the theoretical support of the empirical application.
previous mentioned models are combined [25–29] and other mod- Third, a real data application is executed involving the proposal
els [30–34]. All of these models try to catch, from the past, infor- of a new index and the computation of classical financial indicators
mation on the future. Often, these models are either focused on in our stochastic framework.
specific time scale forecasting, or synthetic time series generation.
In a previous series of papers [35–37], we applied different 2. Research methodology and materials
semi-Markov models to wind speed modeling and we demon-
strated that the semi-Markov framework out performs the The research methodology is summarized in Fig. 1. We use real
Markov models and should therefore be preferred in the modeling wind speed data, described here below, to set the stochastic model
of wind speed. that is described in the next section. Then, through Monte Carlo
The approach we propose here is based on an indexed simulations, we generate synthetic time series of wind speed.
semi-Markov chain (ISMC) model that we advanced in [38] and Both time series, real data and synthetic data, are converted into
applied to the generation of synthetic wind speed time series. In energy by means of the power curve of the chosen wind turbine.
[38] we demonstrated that the ISMC model is able to accurately We then suppose that the produced energy is sold in the market
reproduce the statistical behavior of wind speed. The ISMC model at two different prices (with and without government incentives).
is a non-parametric model because it does not require any assump- The sale produces cash flows on which we compute two important
tion on the form of the distribution function of wind speed. financial indicators, namely, Semi-Elasticity and Relative
Furthermore, we showed that it is able to forecast wind speed at Convexity. To the best of our knowledge, we are the first to use
different time scales without loosing the goodness of forecasting and apply these financial indicators to wind energy investments.
which is almost independent from the time horizon [39]. For the analysis conducted in this paper we use a free database
The main purpose of this paper is to present a profitability anal- of a weather station situated in Italy at N45°280 14,900 –E9°220 19,900
ysis of wind energy production through economical and financial and at 107 m of altitude. The data of the L.S.I. -Lastem station are
indexes. downloadable at www.lsi-lastem.it/meteo/page/dwnldata.aspx.
We beginning by presenting a comparison between the energy The database is composed of more than 230,000 data with a sam-
produced by real wind speed and that obtained by implementing pling frequency of 10 min. The period of sampling starts from
the ISMC model through Monte Carlo simulation. Then, we 25/10/2006 until 28/06/2011, almost five years. The station uses
advance a new indicator that quantifies the utilization degree of a combined speed-direction anemometer at 22 m above the
a wind turbine. This indicator is defined as the Satisfying Power ground. Since we need to test the model for the energy production
Demand Percentage (SPDP) and measures the fraction of time in of a commercial wind turbine, we transpose our database at the
which the wind turbine satisfies the electrical power demand. It turbine rotor altitude using a well known relation, see e.g. [40]:
can be interpreted as a degree of utilization of the wind turbine. a
We compute these indicators for real and synthetic data. Finally, h 1
v h ¼ v rif a¼ ð1Þ
we consider two financial indicators of the cash flow, namely, hrif ln zh0
Semi-Elasticity and Relative Convexity. We compute these indica-
tors on the cash flow generated by selling the electricity produced where v h is the wind speed at the height h of the wind turbine hub
by a wind turbine. The cash flow sequences are unknown and ran- and v rif is the value of the wind speed at the height hrif of the instru-
dom because they depend on the future values of wind speed. ment. In our application h = 50 m and hrif ¼ 22 m. The symbol z0
Furthermore, the cash flows are subject to the interest rate risk. denotes a parameter that takes into account the morphology of
This is an important component that affects the operation risk of the area near the wind turbine. For a region without buildings or
0.4
3. The indexed semi-Markov chain model
0.3
The general formulation of the ISMC model has been developed
Frequency
6 k¼0
time spent in the wind speed i equal to t and the value of the index through Monte Carlo simulation, 1000 possible scenarios of wind
process equal to u. speed time series of length 15 years for the first case and 30 years
for the second case. The sampling period is 10 min.
4. Economic application
4.1. Energy production
Mathematical models of wind speed can be used to assess the
economic validity of an investment in a wind plant. In this section, According to the power curve described in Fig. 3 and Table 1,
we show how the ISMC model can be used to evaluate useful eco- wind speed has been transformed into power and then into pro-
nomic indicators like the Internal Rate of Return, the duced energy in a specific time window. Table 4 displays the
Semi-Elasticity and Relative Convexity of the cash flow generated
by an investment in a wind plant at a specific site. We conduct
Table 5
two separate analyzes: in the first, we consider an investment for Mean value, standard deviation, Skewness, Kurtosis and JB test for real and synthetic
a period of 15 years, which corresponds to the period in which gov- energy produced.
ernment incentives on the energy price in Italy are still present. In
Mean STD Skewness Kurtosis JB jB stat p-value
the second case we consider an investment without government
Annual energy
incentives and in this case, being the price of the energy lower than
Real 147.4 42.90 0.331 2.70 Rejected 29 0.001
in the previous case, the period of the investment is settled to be Synthetic 150.5 37.82 0.283 2.65 Rejected 24 0.001
greater than or equal to 30 years. The time period is defined such
that the initial investment is at least repaid by the selling of energy
and the pay back period is reached.
Since the ISMC model is based on discrete wind speeds, in order
to apply it to real data, we selected eight wind speed values (see 0.05
6 7–8 9
7 8–9
8 >9
Table 3
U m processes discretization. 7
Table 4
Frequencies values of the electrical energy produced in 10 min. Comparison between real and simulated data.
comparison between the distribution of the produced energy in a until 32 m/s, in which the wind turbine produces energy at its
10 min interval for real and synthetic data. Although the transfor- rated power. This explains the existence of the two modes.
mation from wind speed into energy is nonlinear, the ISMC model In Table 5 we give some statistics of the annual energy distribu-
is able to accurately reproduce a real case of energy production. In tion for real and synthetic data. We included the results for the
fact, the values obtained in the two cases are almost identical. The JarqueBera test (JB), which is a goodness-of-fit test that determines
particular form of these distributions is due to the existence of a whether sample data have the skewness and kurtosis matching a
cut-in wind speed, under which no energy is produced, and a part normal distribution. The test rejects the normality hypothesis for
of the power curve that remains at a constant value, from 10 m/s the annual energy. This is mainly due to the absence of a
Gaussian tail of the distribution on its right side (see the distribu-
tion figure in [41]).
17 In Fig. 4, we show the synthetic annual energy distribution
Synthetic Data
Real Data against the real annual energy distribution. Dots close to the
16
straight line demonstrate the appropriateness of the ISMC model.
The coefficient of determination for this set of data is r2 ¼ 0:89.
15
SPDP [%]
Fig. 6. Variation of the SPDP for a 5 years time period as a function of the rated where the numerator represents the period in which the electrical
power over dimensioning. power available (Pa ) satisfies the electrical power required (P r )
Bootstrap ISMC
0.12 0.12
0.08 0.08
Frequency
Frequency
0.04 0.04
0 0
−6.805 −6.795 −6.785 −6.9 −6.8 −6.7
Semi−Elasticity Semi−Elasticity
(a)
Bootstrap ISMC
0.12 0.12
0.08 0.08
Frequency
Frequency
0.04 0.04
0 0
−7.03 −7.02 −7.01 −7.1 −7.05 −7
Semi−Elasticity Semi−Elasticity
(b)
Fig. 7. Histograms of the Semi-Elasticity for real and simulated data. (a) Case with government incentives and cash flow of 15 years. (b) Case without government incentives
and cash flow of 30 years.
G. D’Amico et al. / Applied Energy 154 (2015) 290–297 295
and T is the number of unit of time where the inequality Pa Pr P 0 and we assume to extend the rated power of the wind turbine from
is satisfied in the considered time horizon. To compute this indica- 0% to 100% of its value. For example, if we have a wind turbine with
tor we evaluate the power available (practically the power available a rated power of 10 kW and we want to improve the plant, we are
on the wind turbine) and we compare it against a scaled electrical interested in knowing how the SPDP will change. If we assume a
power demand. We use the data of [42] and scale its maximum 40% of over-dimension (then we will have a turbine of
value. Particularly in [42], the greatest value (during the day) of 10 + 4 kW), the SPDP will be 13%, with a percentage variation of
the power demand is 14.8 MW. We scaled the entire curve to set about 10%.
the maximum value equal to the rated power of the chosen com- It is interesting to note that the variation of the SPDP is very
mercial wind turbine, which is 9.8 kW. The result of the scaling pro- important, it increases almost of 50% with a variation of 100% of
cedure is showed in Fig. 5. This represents the power demand in a the over dimensioning. This behavior is well kept by the model.
day, we extend it for the considered period (five years) without con-
sidering any seasonality. 4.2. Financial indicators
In Fig. 6 we show the comparison of the SPDP evaluated on real
and synthetic data for a period of five years. Moreover, we compute In this subsection we will compare important financial indica-
it by varying the over-dimensioning of the plant. In other words, tors computed on real time series, generated through the boot-
we fix the maximum power required during the day at 9.8 kW strapping procedure, and simulated data, generated through the
Table 6
Statistics of Semi-Elasticity distributions evaluated for real and synthetic data and for the two cases: with government incentives (15 years) and without (30 years).
Bootstrap ISMC
0.12 0.12
0.08 0.08
Frequency
Frequency
0.04 0.04
0 0
−10.53 −10.52 −10.51 −10.6 −10.4 −10.2
0.08 0.08
Frequency
Frequency
0.04 0.04
0 0
−17.98 −17.97 −17.96 −18.05 −18 −17.95
Relative Convexity Relative Convexity
(b)
Fig. 8. Histograms of the Relative Convexity for real and simulated data. (a) Case with government incentives and cash flow of 15 years. (b) Case without government
incentives and cash flow of 30 years.
296 G. D’Amico et al. / Applied Energy 154 (2015) 290–297
Table 7
Statistics of Relative Convexity distributions evaluated for real and synthetic data and for the two cases: with government incentives (15 years) and without (30 years).
ISMC model by means of Monte Carlo simulations. As explained Another important index is the Relative Convexity (RC) that
above, the bootstrapping procedure allows generating model inde- expresses the convexity of the discounted value of the cash flow
pendent time series by randomly resampling the original data. We stream as units of value variation. In formula we have:
consider two different cases: In the first case we compute the cash Pn
flow obtained by selling the energy produced by the wind turbine, 1 ðs þ 1Þ CF s ð1 þ rÞs
s¼1 s C
RC ¼ Pn s ¼ ð6Þ
taking into account Italian government incentives on the selling Se s¼1 CF s ð1 þ rÞ Se
price. This choice is motivated by interest determine whether the
investment can also be justified without the government incen- and C denotes the convexity of the cash flow. The behavior of the
tives. In Italy, for a wind plant with a rated power under 1 MW duration and of the convexity were computed, by the same authors,
and until 15 years after its setup, the price of the energy is fixed in [41].
(at the time of writing, to 0.291 €/kW h). In this case we consider In Fig. 8 we plotted the Relative Convexity in the same cases as
a cash flow of 15 years. Just to give a quantitative comparison, the previous indicators and in Table 7 its statistic values are
the second case takes into account the selling of energy at the pre- summarized.
sent market price in Italy, without government incentives As can be observed, for each pair of indicators and for each
(0.027 €/kW h). In this case the considered period of the invest- specific investment period, the mean values of the indicators are
ment is extended to 30 years. almost the same for real and simulated data.
For each case we generate 1000 possible scenarios and we eval-
uate the financial indicators for both real and synthetic data. The
5. Conclusion
trajectories are transformed into power through the power curve
of the wind chosen turbine, then into energy by considering the
We defined the Satisfying Power Demand Percentage as a new
time between two successive steps and after in Euro by multiply-
indicator that quantifies the utilization degree of a wind turbine
ing for the specific price considered. The annual cash flows are
and we evaluated it in conjunction with two financial indicators
built and, at time 0, the cost of the initial investment is considered.
(Semi-Elasticity and Relative Convexity)) of the cash flow gener-
Commonly for wind turbines with a rated power under 100 kW, a
ated by the sale of electrical energy produced by a given wind tur-
price of 3000 € for each kW is considered (this value comes from
bine. These indicators provide valuable information concerning the
some quotes obtained from sector companies). This price takes into
suitability of investment in a wind farm and measure the degree of
account the cost of the turbine, the setup cost and also the annual
risk involved in wind energy production in conjunction with the
maintenance cost. Given that our wind turbine has 10 kW of rated
interest rate risk.
power, we chose the initial investment of 30,000 €.
The computations were performed on simulated data obtained
The considered indicators are the Semi-Elasticity and Relative
by an indexed semi-Markov chain (ISMC) model of wind speed and
Convexity.
compared with those evaluated on real data. The results suggest
that the ISMC model is a correct model for wind speed analysis
and applications.
4.2.1. Indexes used for the measurement of interest rate risk Future work will focus on using the evaluation of the proposed
In this subsection we evaluate some measures of the interest indicators to evaluate new wind farm sites by applying them to
rate risk that are commonly used in financial analyzes and we real case of investments in wind energy production and to extend
apply them to the investment in a wind plant. As we have seen, the analysis when we assume a stochastic evolution of the interest
the production of energy generates a cash flow sequence. The value rate process.
of this stream may change in time due to variations of the interest The importance of the present work is twofold: from one side
rate. The wind energy producer, as any other firm, may suffer the we have introduced indicators that can be used during the decision
risk of changes in the interest rate and then it is necessary to con- process of investing in a wind farm in a specific site; on the other
trol such an uncertainty. site we have showed that the ISMC model reproduces, almost
One of the most important indexes is the Duration (D). It is exactly, the results obtained with real wind data, meaning that
defined as a weighted average of time to payment of the cash flow the ISMC model is well suited for producing synthetic data of wind
sequence. From D it is possible to recover the Semi-Elasticity (Se), speed.
which is a relative measure of the variation of the value of the cash
flow stream. Depicted in formula we have:
Acknowledgement
Pn s
1 s¼1 s CF s ð1 þ rÞ D
Se ¼ P n s ¼ : ð5Þ The comments and helpful suggestions made by two anony-
1þr CF s ð1 þ rÞ 1 þr
s¼1 mous referees are gratefully acknowledged.
[2] Suomalainen K, Pritchard G, Sharp B, Yuan Z, Zakeri G. Correlation analysis on [23] Hocaoğlu FO, Gerek ÖN, Kurban M. A novel wind speed modeling approach
wind and hydro resources with electricity demand and prices in New Zealand. using atmospheric pressure observations and hidden markov models. J Wind
Appl Energy 2014;137:445–62. Eng Ind Aerodynam 2010;98(8):472–81.
[3] Sahin C, Shahidehpour M, Erkmen I. Generation risk assessment in volatile [24] Fawcett L, Walshaw D. Markov chain models for extreme wind speeds.
conditions with wind, hydro, and natural gas units. Appl Energy 2012;96:4–11. Environmetrics 2006;17(8):795–809.
[4] Johnston L, Díaz-González F, Gomis-Bellmunt O, Corchero-García C, Cruz- [25] Pourmousavi Kani S, Ardehali M. Very short-term wind speed prediction: a
Zambrano M. Methodology for the economic optimisation of energy storage new artificial neural network-markov chain model. Energy Convers Manage
systems for frequency support in wind power plants. Appl Energy 2011;52(1):738–45.
2015;137:660–9. [26] Denison D, Dellaportas P, Mallick B. Wind speed prediction in a complex
[5] Chang TP. Estimation of wind energy potential using different probability terrain. Environmetrics 2001;12(6):499–515.
density functions. Appl Energy 2011;88(5):1848–56. [27] Pandey M, Van Gelder P, Vrijling J. Bootstrap simulations for evaluating the
[6] Chang TP. Performance comparison of six numerical methods in estimating uncertainty associated with peaks-over-threshold estimates of extreme wind
weibull parameters for wind energy application. Appl Energy velocity. Environmetrics 2003;14(1):27–43.
2011;88(1):272–82. [28] Liu H, Tian H-Q, Chen C, Li Y-f. A hybrid statistical method to predict wind
[7] Nagai BM, Ameku K, Roy JN. Performance of a 3 kw wind turbine generator speed and wind power. Renew Energy 2010;35(8):1857–61.
with variable pitch control system. Appl Energy 2009;86(9):1774–82. [29] Lei M, Shiyan L, Chuanwen J, Hongling L, Yan Z. A review on the forecasting of
[8] Danao LA, Eboibi O, Howell R. An experimental investigation into the influence wind speed and generated power. Renew Sustain Energy Rev
of unsteady wind on the performance of a vertical axis wind turbine. Appl 2009;13(4):915–20.
Energy 2013;107:403–11. [30] Salcedo-Sanz S, Perez-Bellido AM, Ortiz-García EG, Portilla-Figueras A, Prieto L,
[9] Lenzen M. Life cycle energy and greenhouse gas emissions of nuclear energy: a Paredes D. Hybridizing the fifth generation mesoscale model with artificial
review. Energy Convers Manage 2008;49(8):2178–99. neural networks for short-term wind speed prediction. Renew Energy
[10] Davidsson S, Höök M, Wall G. A review of life cycle assessments on wind 2009;34(6):1451–7.
energy systems. Int J Life Cycle Assess 2012;17(6):729–42. [31] Segura-Heras I, Escrivá-Escrivá G, Alcázar-Ortega M. Wind farm electrical
[11] Zimmermann T, Gößling-Reisemann S. Influence of site specific parameters on power production model for load flow analysis. Renew Energy
environmental performance of wind energy converters. Energy Procedia 2011;36(3):1008–13.
2012;20:402–13. [32] Tar K, Szegedi S. A statistical model for estimating electricity produced by
[12] Zimmermann T. Parameterized tool for site specific lcas of wind energy wind energy. Renew Energy 2011;36(2):823–8.
converters. Int J Life Cycle Assess 2013;18(1):49–60. [33] Lazić L, Pejanović G, Živković M. Wind forecasts for wind power generation
[13] Jiang Y, Song Z, Kusiak A. Very short-term wind speed forecasting with using the eta model. Renew Energy 2010;35(6):1236–43.
bayesian structural break model. Renew Energy 2013;50:637–47. [34] Abdel-Aal R, Elhadidy M, Shaahid S. Modeling and forecasting the mean hourly
[14] Liu H, Tian H-q, Li Y-f. Comparison of two new arima-ann and arima-kalman wind speed time series using gmdh-based abductive networks. Renew Energy
hybrid methods for wind speed prediction. Appl Energy 2012;98:415–24. 2009;34(7):1686–99.
[15] Cassola F, Burlando M. Wind speed and wind energy forecast through kalman [35] D’Amico G, Petroni F, Prattico F. First and second order semi-markov chains for
filtering of numerical weather prediction model output. Appl Energy wind speed modeling. Physica A: Stat Mechan Appl 2013;392:1194–201.
2012;99:154–66. [36] D’Amico G, Petroni F, Prattico F. Performance analysis of second order semi-
[16] Bivona S, Bonanno G, Burlon R, Gurrera D, Leone C. Stochastic models for wind markov chains: an application to wind energy production. Method Comput
speed forecasting. Energy Convers Manage 2011;52(2):1157–65. Appl Probability 2014:1–14.
[17] Poggi P, Muselli M, Notton G, Cristofari C, Louche A. Forecasting and simulating [37] D’Amico G, Petroni F, Prattico F. Reliability measures of second-order semi-
wind speed in corsica by using an autoregressive model. Energy Convers markov chain applied to wind energy production. J Renew Energy 2013. Article
Manage 2003;44(20):3177–96. ID 368940.
[18] Ailliot P, Monbet V, Prevosto M. An autoregressive model with time-varying [38] D’Amico G, Petroni F, Prattico F. Wind speed modeled as an indexed semi-
coefficients for wind fields. Environmetrics 2006;17(2):107–17. Markov process. Environmetrics 2013;24:367–76.
[19] Shamshad A, Bawadi M, Wan Hussin W, Majid T, Sanusi S. First and second [39] D’Amico G, Petroni F, Prattico F. Wind speed and energy forecasting at
order markov chain models for synthetic generation of wind speed time series. different time scales: a nonparametric approach. Physica A: Stat Mechan Appl
Energy 2005;30(5):693–708. 2014;406:59–66.
[20] Song Z, Jiang Y, Zhang Z. Short-term wind speed forecasting with markov- [40] Gass V, Strauss F, Schmidt J, Schmid E. Assessing the effect of wind power
switching model. Appl Energy 2014;130:103–12. uncertainty on profitability. Renew Sustain Energy Rev 2011;15(6):2677–83.
[21] Kantz H, Holstein D, Ragwitz M, Vitanov NK. Markov chain model for turbulent [41] D’Amico G, Petroni F, Prattico F. Forecasting wind speed financial return, arXiv
wind speed data. Physica A: Stat Mechan Appl 2004;342(1):315–21. preprint arXiv:1312.3895; 2013.
[22] Castino F, Festa R, Ratto C. Stochastic modelling of wind velocities time series. J [42] Mabel MC, Raj RE, Fernandez E. Analysis on reliability aspects of wind power.
Wind Eng Ind Aerodynam 1998;74:141–51. Renew Sustain Energy Rev 2011;15(2):1210–6.