Vous êtes sur la page 1sur 9

158

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 3, NO. 1, JANUARY 2012

Wind Power Projects Planning Considering Real Options for the Wind Resource Assessment
Eduardo Alejandro Martinez-Cesena, Student Member, IEEE, and Joseph Mutale, Senior Member, IEEE
AbstractInvestments in wind power projects (WPPs) have increased in the last few years. This trend is partially due to the availability of support schemes, which increase the economic attractiveness of WPPs. Alternatively, the value of WPPs can be enhanced by improving available techniques used for their planning and design. After reviewing WPP literature, it was concluded that available tools for the planning and design of WPP could be improved by addressing the uncertainty of the wind resource assessment (WRA), and this source of uncertainty could be used to enhance the value of WPPs with real options (ROs) theory. ROs theory is known for its potential to increase the expected worth of projects by exploiting the value of exibility within the projects investment decisions and designs. Nevertheless, ROs literature has to be extended to properly address the design of WPPs. Based on the gaps in ROs theory and WPPs planning, this paper proposes a methodology that relies on ROs theory to incorporate WRA uncertainty in the planning and design process of WPPs. The methodology is illustrated with a small case study and its potential to increase the value of WPPs under different conditions is analyzed for a wide range of case studies. The results illustrate the circumstances and assumptions that can improve and weaken the effectiveness of the methodology. It is concluded that the application of the proposed ROs methodology results in increased value for WPPs in most scenarios. Index TermsGeneration planning, real options (ROs), renewable energy, wind resource assessment (WRA).

I. INTRODUCTION N the last few decades, increasing electricity power demand and environmental concerns worldwide have driven investments in electricity generation projects based on renewable energy sources. Among such projects, wind power projects (WPPs) have had one of the highest rates of growth [1]. Nevertheless, the success of WPPs is partially due to the implementation of support mechanisms such as feed-in tariffs and carbon trading markets, which improve the economic attractiveness of WPPs [2]. Alternatively, the nancial appeal of WPPs can also be enhanced by researching and developing wind technologies, and methodologies used for the planning and design of WPPs. The latter is the focus of this paper.

Manuscript received March 31, 2011; revised July 23, 2011; accepted July 30, 2011. Date of publication August 08, 2011; date of current version December 16, 2011. This work was supported in part by the Mexican Council of Science and Technology (CONACYT) under Grant 305445. The authors are with the University of Manchester, School of Electrical and Electronic Engineering, Electrical Energy and Power Systems Group, Manchester, M60 1QD, U.K. (e-mail: Eduardo.Martinez-Cesena@ postgrad.manchester.ac.uk; j.mutale@manchester.ac.uk). Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TSTE.2011.2164102

The planning and design of WPPs is not a trivial matter. It involves the identication of sites with adequate wind resources, selection of turbine models and hub height, formulation of turbine layouts (micrositing), and other tasks. The selection of candidate wind sites and proper characterization of the wind resource is of prime importance for WPPs because a 1% error in the estimation of the wind resource can lead to around a 3% error in the power production of the system [3]. The process to characterize and forecast the wind resource is called wind resource assessment (WRA). The most reliable and typical methods used for WRAs are based on on-site and reference wind data. Such methods can provide acceptable estimate of the wind resource with just one year of on-site measurements, but the assessment can be extended to two or three years to increase the accuracy of the WRA [1], [4]. However, the convenience of extending the WRA is not easily determined because it results in additional costs related to land lease and assessment procedures, and there is no guarantee the estimation of the wind resource will improve. The design of WPPs can be optimized considering prots maximization, costs minimization, and other nancial criteria; nancial, technical, and environmental constraints; and several sources of uncertainty, especially the uncertainty associated to the WRA. For such purpose, several approaches, ranging from mathematical programming to meta-heuristic techniques, have been proposed to optimize the planning and design of WPPs [3], [5], [6]. Available approaches for the planning and design of WPPs assume the wind prole is stochastic but the distribution of the wind speeds are xed and known with certainty. Roy [7], [8] presented two methodologies for the planning of WPPs considering different available wind turbine models and hub heights. Wind speed is modeled as a xed data series. Serrano-Gonzalez et al. [9], [10] proposed several genetic algorithms for the design of WPPs considering micrositing as an optimization variable. Wind speed is assumed to follow a Weibull distribution with known shape and scale parameters. Banzo and Ramos [11] propose a tool for the planning and design of offshore wind farms considering stochastic wind speeds and system reliability. Wind speed is modeled using Rayleigh distributions with xed mean speed. This WPP literature suggests that current methodologies for the planning and design of WPPs address the variability of the wind resource, but assume the distribution of the wind resource is known, which neglects uncertainty in the WRA. Several reasons to neglect WRA uncertainty might include the considerations that current WRA techniques are highly accurate and can produce power generation probability density functions based on forecast uncertainty, as the methods proposed in [12] and [13], and that including the forecast uncertainty in the planning and design process of WPPs would com-

1949-3029/$26.00 2011 IEEE

MARTINEZ-CESENA AND MUTALE: WPPs PLANNING CONSIDERING ROs FOR THE WIND RESOURCE ASSESSMENT

159

plicate the process and likely offer trivial benets. Nevertheless, even if the WRA technique is deemed accurate and can produce power generation probability density functions, the performance of the WRAs is case specic and dependent on several factors such as the characteristics of the site and amount of data available. Additionally, the performance of the WRA is likely to improve if data derived from additional on-site measurements becomes available and, as will be shown in this paper, computational difculties derived from addressing WRA uncertainty are not signicant and the resulting benets can be substantial depending on the characteristics of the WPPs. The potential benets WPP owners could accrue from considering WRA uncertainty in the planning and design of the projects can be more fully and adequately evaluated using real options (ROs) theory. ROs theory is known for its capacity to enhance the value of projects under uncertainty, and theoretically can be used for WPPs under WRA uncertainty. However, ROs literature addressing the design of projects is scarce and the literature has to be extended to address the design of WPPs. Accordingly, this paper proposes an ROs methodology for the planning and design of WPPs that complements existing tools by incorporating WRA uncertainty in the design of projects to estimate optimal investment timing. The rationale behind the methodology is that project planers posses options, but not obligations, to delay construction of WPPs with the objective of collecting additional on-site measurements in expectance that the WRA would become more accurate and allow better designs, and thus more valuable projects. The application of the proposed methodology is illustrated with a small example, and the potential of the methodology to enhance the value of WPPs is shown using several case studies and sensitivity analyses. The results show that the methodology increases the expected value of WPPs in most cases, and illustrate the circumstances that improve and weaken the performance of the methodology. The rest of the paper is structured as follows. In Section II, ROs theory is described, and its current development and adequacy for this research are discussed. Afterwards, in Section III, the ROs methodology for the planning and design of WPPs considering WRA uncertainty is presented. Later, in Section IV, the application of the ROs methodology is illustrated with a small case study. Then, in Section V, multiple case studies are used to show the accuracy of the methodology to identify optimal investment timing and the value of ROs under different circumstances and assumptions. Finally, in Section VI, the general conclusions and main ndings of this work are discussed. II. REAL OPTIONS THEORY ROs theory is the application of nancial options theory for planning and assessing projects concerning real assets. According to this theory, project planers possess options, but not obligations, to modify projects in response to the evolution of uncertainty. The identication and proper use of these options such as rushing, delaying, abandoning, or adjusting investment decisions, provides exibility to projects and potentially increases their expected value and decreases their risks [14]. ROs theory was introduced as a new research area in the late 1970s due to growing discontent with available techniques for

the assessment of projects under uncertainty [15][17], and signicant innovations in options pricing techniques such as the BlackScholes formula [18] and, later on, the binomial lattice approach [19]. At the time, it became evident that available discounted cash ows techniques tend to undervalue projects under uncertainty because they fail to address the exibility project managers possess to adjust projects. It was necessary to improve available methods to incorporate exibility, and this could be achieved by modeling exibility as options to adjust projects in response to uncertainty. Nevertheless, assessing which options were convenient for the project was complicated due to the different costs and uncertain benets offered by each option. This problem was initially solved with the introduction of the BlackScholes equation, which provided a straightforward means to determine the value of options. ROs theory has been applied to many different types of projects, including power generation projects [20][25], and as a result several techniques for the assessment of ROs have been developed [14], [17]. This would suggest that identifying an available ROs technique suitable to incorporate WRA uncertainty in the planning and design of WPPs is a simple task. Nevertheless, the majority of ROs literature neglects the design of projects to focus on options derived from project management [26], [27], which makes the adequacy of most available ROs techniques to address the design of WPPs questionable. Available ROs literature considering the design of systems is scarce, but it suggests that the application of ROs theory considering the specic characteristics of projects allows the identication of better project designs [28][30]. However, the resulting ROs tend to be complex, interdependent, and path-dependent [27], [31]. These characteristics must be addressed by the ROs technique selected for this study. Available techniques for ROs assessment can be classied as those based on partial differential equations, such as the BlackScholes equation; trees and lattice processes, such as the binomial lattice; and simulation techniques, such as Monte Carlo simulation [17], [32]. Partial differential equations allow the valuation of options in a fast and straightforward manner. However, formulating adequate equations and justifying assumptions might be signicantly difcult [17]. As an example, the no-arbitrage and geometric Brownian assumptions used in the BlackScholes equation cannot be justied for ROs that impact the design of projects [33]. Trees and lattice approaches are more robust, allow modeling different types of ROs, and path-dependency can be considered when formulating the trees [27]. On the other hand, accurate results might require the formulation of trees of high order with multiple time periods, which increases computational burden. Simulation techniques can handle different stochastic processes but, as forward-moving techniques, they are not adequate to model options with possible early exercises and their use can be computationally expensive [17]. Considering the characteristics of available ROs techniques, an ROs methodology based on path-dependent scenario trees and simulations was proposed to assess options in generic engineering systems in [27]. The methodology was applied to satellite and hydropower systems. Afterwards, the methodology was modied to exploit some of its areas of improvement in [26].

160

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 3, NO. 1, JANUARY 2012

Fig. 2. Wind forecast standard deviation evolution.

Fig. 1. Proposed ROs methodology.

Accordingly, a similar ROs technique based on scenario trees and simulations is developed for this work. The proposal of this work is that, from an ROs perspective, project planners have options to extend the WRAs period in the expectation that more accurate wind resource estimations lead to more valuable WPPs. Such an option is assessed using an ROs technique to properly address its costs and uncertain benets, and the ROs technique is based on scenario trees and simulations in order to address the design of WPPs. The result is an ROs methodology that complements current tools for the planning and design of WPPs by including benets derived from addressing WRA uncertainty. III. METHODOLOGY The proposed ROs methodology can be used by WPPs planners once a WRA is available to determine if 1) a WPP should be designed using data from the available WRA, 2) the project should be abandoned, or 3) the investment decision should be deferred for a time period to extend the WRA, and the methodology must be used again in the next time period when the WRA is updated. The methodology must be performed every time the WRA is updated after delaying the WPP, and until an investment decision to either abandon or build the WPP is achieved. As a result, the proposed approach should lead to optimal investment timing in WPPs, which is the time period when the accuracy of the WRA allows the construction of the most valuable WPP according to an objective function. Based on the case studies considered in this work, the WRA is typically extended less than ve years. Literature suggests that the WRA can usually be extended two or three years [1], [4]. The proposed ROs methodology can be divided into ve stages, as shown in Fig. 1. A. Error Forecast At the rst stage of the methodology, the expected error of the available WRA and its expected error if extended by a specied

time period are estimated using simulations. The possible evolution of the WRA error is the basis to determine the potential benets of an RO to delay the project. The rate by which the expected error of the WRA decreases with longer measurement periods is mainly a function of the forecasting method (i.e., the methods previously mentioned [12], [13]), standard deviation of on-site wind speeds, and correlation between on-site and reference data. The performance of different WRA techniques can be obtained from the literature [34], the standard deviation of wind speeds at a site can be approximated using wind speed prediction methods such as atlases and simulation models [3], and the correlation between on-site and reference data can be estimated with concurrent on-site and reference data. The models used for the error forecast simulations must represent the particular characteristics of the wind site and WRA technique used for the planning and design of a WPP. In this work, wind speeds are simulated assuming Weibull distributions for wind speeds throughout the years and Gaussian distributions for annual wind speed variations [1], [35]. Considering the literature recommends the use of complete years of on-site measurements for wind speed forecasts to avoid seasonal variations [1] and the accuracy of the WRA is a function of the standard deviation and correlation, the long-term wind speed is forecasted with [35] (1) where and are, respectively, the forecasted and measured on-site annual mean wind speeds, and are, respectively, the long-term and measured reference annual mean wind speeds, is the correlation between on-site and reference annual mean wind speeds, and and are the on-site and reference annual mean wind speed standard deviations, respectively. Monte Carlo simulations (10 000 scenarios) are used to generate data series with the expected error in the long-term wind speed forecast considering different lengths of on-site measurements. An example of the expected forecast error according to (1), different correlation values , and a 6% annual wind speed standard deviation is shown in Fig. 2. The outputs of this stage are data series with the expected evolution of the error in the long-term wind speed. Only data corresponding to two time periods, the current and next time periods, are used.

MARTINEZ-CESENA AND MUTALE: WPPs PLANNING CONSIDERING ROs FOR THE WIND RESOURCE ASSESSMENT

161

Fig. 3. Linked scenario trees.

B. Scenario Trees Formulation In this stage, linked scenario trees are formulated to depict the expected error in the available and extended WRA. The idea is to identify the most likely values of the long-term wind speed considering the error in the WRA. As previously discussed, ROs literature recommends the use of path-dependent scenario trees to assess options concerning the designs of systems [26], [27]. Nevertheless, this was proposed considering 1) the present is known with certainty and the future is uncertain, which justies the use of a single tree with multiple branches depicting possible future scenarios; and 2) the availability of ROs is affected by previous actions, which can only be modeled with path-dependent trees. Such assumptions do not hold for this work where 1) the uncertainty of the WRA is the highest in the present and it will decrease in the future, and 2) the methodology is meant to be used every time the WRA is updated, which allows the ROs assessment to focus on analyzing the present and next time periods and break the necessity of modeling path-dependency. Based on these factors, it is proposed to begin the ROs assessment with multiple linked scenario trees considering just two time periods and, later on, group the nodes of the scenario trees to create a single decisions tree, which is used to estimate the value of ROs under realistic scenarios. The linked scenario trees can be formulated by binning the data series produced in the previous stage of the methodology and assuming the available WRA is accurate. The data of the available WRA is binned in scenarios with the same probability of occurrence. The data of the extended WRA is binned based on the corresponding scenario of the data of the available WRA, and the behavior (increase or decrease) of the magnitude of the forecasted long-term wind speed. An example considering four linked binomial trees is shown in Fig. 3. It should be noticed that most or all wind speeds in the scenario trees differ from the wind speed estimated in the available WRA. The differences between these wind speeds represent the expected error in the wind speed forecast. C. Scenario Trees Assessment At this stage, several designs are formulated for the WPP considering the different speeds in the scenario trees. Afterwards, each design is assessed based on the long-term wind speed provided by the available WRA. The idea is to model the fact that errors in the long-term wind speed forecast can lead to suboptimal WPP designs.

The design of the WPPs can be optimized with available methodologies, such as the approaches previously discussed [7][11], and considering several sources of uncertainty such as the price of electricity or the availability of renewable energy support schemes. In this work, the design of WPPs is optimized assuming the micrositing, turbine model, and hub height are the only optimization variables and WRA uncertainty is the sole source of uncertainty. The mathematical models involved in the optimization procedure include 1) Weibull probability density functions for wind speed distributions [36], 2) wind shear models based on the friction coefcient of the terrain to determine wind speeds at different heights [36], 3) linear, parabolic, and cubic models to approximate the wind turbines power curves [37], 4) wind wake interaction models to determine the efciency of different turbine layouts [38], and 5) an objective function, which in this work is the maximization of the net present value (NPV) as dened by [36] (2a) (2b) (2c) where is the cost of the WPP, , , , and are the costs due to the WRA, initial investment, land lease costs, and operation and maintenance, respectively; is the NPV of the WPP, is the price of electricity, is the generation of the WPP, and the subscript represents time periods (years). It is assumed that the WPPs perceive benets solely due to energy sold to the grid at a xed price. The assumption that the price of electricity is deterministic, is a typical consideration in literature regarding the planning and design of WPPs [5], [8][10]. However, in practice, this assumption would not hold if wind power is traded in a highly volatile electricity market. In such a case, neglecting the electricity price uncertainty could lead to under or over estimations of the projects benets. In this work, the price of electricity is varied to simulate changes in the protability of WPPs, such as the changes resulting from the uncertainty of electricity prices. Nevertheless, the electricity price uncertainty is not further addressed because it is considered beyond the scope of this work.1 The costs of the project comprise cost due to the WRA, initial investment, land lease, and operation and maintenance. The WRA costs include costs due to acquisition of measurement equipment and reference data, and monitoring. The initial investment includes costs related to the acquisition and installation of the wind turbines, civil works, and electrical infrastructure, among others. Land lease costs include periodic costs for the use of land. The costs of operation and maintenance comprise costs due to monitoring, maintenance, and repairs of the WPPs.
1A detailed explanation of the effects of electricity price uncertainty in the planning and design of electricity generation projects, and the application of ROs theory to address such a source of uncertainty can be seen in one of the authors previous works [26].

162

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 3, NO. 1, JANUARY 2012

TABLE I WIND ROSE DATA

Fig. 4. Scenario trees and corresponding decisions tree.

The design of the WPPs is optimized using exhaustive searches [39] and simulations implemented in MATLAB. D. Decisions Tree At this stage, the nodes of the scenario trees are grouped to create a decisions tree, which can be used to assess an RO to delay the WPP for a time period or make investment decisions using the available WRA. The nodes of the scenario trees cannot be used to assess investment decisions because they do not depict realistic investment decision scenarios. In other words, if investment decisions are associated to the scenario trees, the analysis would suggest postponing investment decisions when the WRA is inaccurate and vice versa, which is unrealistic because it implies the project planners know the actual error in the available WRA. Project planners can estimate the expected error range in the WRA and can monitor the changes in the wind speed forecast. Based on this, realistic investment decisions can be associated to groups of nodes of the scenario trees whenever the nodes are related to the same time period and same variation of the wind speed forecast (increase or decrease). An example of such a tree is shown in Fig. 4; each group of nodes is identied by a color (white, black, and gray). E. Investment Decision The use of decision trees for the optimization process might imply the need of stochastic mathematical programming techniques to assess ROs as was done in [26] and [27]. However, considering the trees formulated by the methodology only include two time periods, the optimization process can be reduced to the following three steps: 1) use (3) to compute the NPV of investment decisions for all decisions nodes, 2) use (4) to compute the NPV of immediate and delayed investments decisions, and 3) select the alternative with the highest NPV (3) (4) is the probability of occurrence of a scewhere nario tree node, is the discounted cost of the WRA, is the NPV of the WPP, and subscripts , , and indicate parameters corresponding to a decision tree node, scenario tree node, and time period, respectively.

If the expected NPV of the project is negative, the methodology would suggest abandoning the project. Otherwise, an investment can be made. If the expected NPV of the WPP in the rst time period is the highest, the methodology recommends an immediate investment and the RO to delay investment decisions has no value. If the expected NPV of the deferred project is the highest, an investment delay is recommended, the methodology should be used again after the WRA has been extended for one time period, and the value of the RO is the difference between the expected NPV of a delayed and an immediate investment decision. IV. ILLUSTRATIVE EXAMPLE The application of the methodology can be illustrated with a small case study. In this example, the proposed ROs methodology is used to analyze investment decisions for a single time period. However, the complete application of the methodology involves applying the ROs methodology every time the WRA is updated after an investment decision delay, until the WPP is either built or abandoned. A WRA based on one year of correlated data is available for a site with wind potential for power generation. The project planners can use the available WRA and the proposed ROs methodology to assess if: 1) the WPP should be designed using the available wind speed data, 2) the project should be abandoned, or 3) the WRA should be extended for a year and consequently investment decisions should be delayed for a year. The available WRA suggests a long-term wind speed of 7.1 m/s at 50 m with a shape parameter of 2.6. The friction coefcient at the site is 0.28. The wind speed in different directions is shown in Table I [40]. The WPP can be designed using wind turbines with a capacity of 1500 kW, diameter of 77 m, hub height of 100 m, and cut-in, rated, and cut-out wind speeds of 4, 12, and 26 m/s, respectively [41]. The cost of buying and installing each turbine is [42]. The WPP is planned based on the maximization of the projects NPV and the following assumptions: 1) the discount factor is 10%, 2) the electricity price is $0.05/kWh, 3) the land lease costs are $57 782/year, 4) the initial and periodic costs of the WRA are $8000 and $24 000/year, respectively [4], and 5) the useful lifetime of the WPP is 20 years. A. Error Forecast At this stage, the expected error of a WRA based on one and two years of data is computed with Monte Carlo simulations with the following assumptions: 1) the mean annual wind speed is normally distributed with a 6% standard deviation, 2) the correlation between on-site and reference wind speeds is 32%, and 3) the WRA is modeled annually using (1). The results indicate

MARTINEZ-CESENA AND MUTALE: WPPs PLANNING CONSIDERING ROs FOR THE WIND RESOURCE ASSESSMENT

163

TABLE II EXPECTED NPV PER DESIGN PER NODE

Fig. 5. Forecast error PDFs.

TABLE III EXPECTED NPV ACCORDING TO THE AVAILABLE FORECAST

Fig. 7. Decision tree for the illustrative study. Fig. 6. Error scenario trees.

E. Investment Decision that extending the WRA for a year would decrease its expected error, as shown in Fig. 5. B. Scenario Trees Formulation The PDFs computed in the previous stage of the methodology are binned to create two binomial scenario trees shown in Fig. 6. According to the scenario trees, the expected mean error in the WRA is 4.5%. The expected error of the WRA if extended for a year can either decrease to 1% with a probability of 76%, or increase to 6.6% with a probability of 24%. C. Scenario Trees Assessment A WPP design is formulated for each node of the scenario tree using the NPV maximization criterion. For simplication, it is assumed that there are only three possible designs for the WPP; each design comprehends a different micrositing and amount of turbines. The expected NPV of WPPs built following each design in each node is shown in Table II. The designs selected for each node are highlighted in bold. The selected WPP designs are assessed again with the NPV criterion, but this time using the available wind speed forecast (7.1 m/s). The results are shown in Table III. D. Decisions Tree The nodes of the scenario trees are grouped in a decision tree as shown in Fig. 7. The value of investing immediately and postponing investment decisions is calculated with (3) and (4). The expected value of an immediate investment decision is

The expected value of postponing investment decisions is

The results show that the expected NPV of the WPP due to an immediate investment is lower than the expected NPV of the project if delayed a year, and the RO to delay the WPP has a value of . Therefore, it is recommended to delay investment decisions for one year. Afterwards, the WRA should be updated and the proposed ROs methodology should be used again. V. CASE STUDY As previously discussed, the proposed methodology has the potential to enhance the expected value of WPPs by incorporating WRA uncertainty in their planning and design using ROs to delay the project, which should lead to optimal investment timing. However, it is important to notice that the value of ROs to delay investment decisions in WPPs and the accuracy of the

164

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 3, NO. 1, JANUARY 2012

TABLE IV ASSUMPTIONS FOR THE CASE STUDY

TABLE V MEAN EXPECTED ACCURACY OF THE ROs METHODOLOGY

methodology to identify optimal investment timing are case specic, and can be affected by underlying assumptions. Accordingly, the performance of the proposed ROs methodology is analyzed using a wide range of case studies and assumptions. The case studies comprise the combination of six wind sites [40], 53 turbine models with variable hub heights [41], and 729 combinations of input parameters for the assessment and design of WPPs described in Table IV. The input parameters were selected considering land costs are proportional to the annual kWh generation potential of each site [42], standard deviations and correlations correspond to those reported in the United States [35], the turbine curves are approximated using models based on the wind turbines cut-in, rated, and cut-out wind speeds [37], the prices of electricity correspond to the costs ranges of WPPs [1], and the discount factor is varied based on conventional values. The WPPs are planned in accordance to the NPV criterion and assuming the wind sites are square with an area of 1 km , the initial and periodic costs of the WRA are $8000 and $24 000/year, respectively [4], and the useful life time of the WPPs considered for the study is 20 years. Each case study addresses one site, one turbine model with a xed hub height, and one combination of parameters taken from Table IV. Moreover, each case study is assessed considering several probabilistic scenarios (500 scenarios per case study) for concurrent yearly on-site and reference wind speeds. The ROs methodology is used repeatedly in every probabilistic scenario until it recommends an investment decision, which is normally achieved in less than ve years. The accuracy of the methodology is calculated as the percentage of events where the ROs methodology recommends investment decisions in the optimal time period. The value of ROs is computed as the mean difference between the NPV of investment decisions corresponding to the time period recommended by the methodology and immediate investment decisions. It is important to notice that although the accuracy of the methodology and the expected value of ROs are related, i.e., the whole value of ROs in a specic case is perceive when accuracy is %100, the magnitude of accuracy and the whole value of ROs are independent of each other. In a case where delaying the project is signicantly inconvenient for the WPP, the ROs methodology is likely to recommend immediate investments in most or all scenarios, which would result in an accuracy of nearly 100% and no value for ROs. On the other hand, in scenarios where ROs are signicantly convenient and the expected error of the WRA is high, the value of ROs would be high and the accuracy of the methodology would be low. The amount of case studies considered in this work is significant, and applying the methodology to asses all scenarios of

TABLE VI MEAN EXPECTED NPV OF THE RO TO DELAY INVESTMENT

each case study can be computationally expensive. In order to reduce computation time, the ROs assessment is based on two monomial error trees and the wind speed simulations are based on Latin hypercube sampling (LHS) [43]. Even though, the use of several multinomial scenario trees can potentially increase the accuracy of the ROs methodology, the results show that two monomial scenario trees sufce to identify optimal investment timing in most scenarios. LHS is proven to provide stable results and reasonably accurate approximation of random distributions with signicantly fewer samples than general Monte Carlo simulations [43]. In order to validate the consistency of the results, the study was repeated ve times; the individual results varied with a 2% or lower standard deviation, which does not affect the general conclusions of this work. The results, namely the accuracy of the methodology and the expected value of ROs are shown, respectively, in Tables V and VI. In order to better explain the results, the effects of the parameters on the accuracy of the methodology and the value of ROs are discussed separately. Land lease costs directly affect the costs of ROs. Increasing land costs decreases the value of ROs due to additional expenses to secure the land when the WPPs are delayed, and vice versa. Results suggest the inuence of this parameter on the accuracy of the methodology is trivial. The correlation and standard deviation of the wind resource directly inuence the accuracy of the WRA. Low correlations and high standard deviations increase the uncertainty of the WRA, which complicates assessing if available WRAs are accurate enough to design the WPP and decreases the accuracy of the methodology. Additionally, when the WRA becomes uncertain, ROs become more valuable. The opposite effects occur for high correlations and low standard deviations. The proposed ROs methodology relies on estimates of the standard deviation and correlation to characterize the expected error of the WRA. Inaccuracies in the estimation of these parameters can affect the performance of the methodology as explored with a sensitivity analysis based on a single case study shown in Fig. 8.

MARTINEZ-CESENA AND MUTALE: WPPs PLANNING CONSIDERING ROs FOR THE WIND RESOURCE ASSESSMENT

165

Fig. 8. Performance of the methodology as a function of the standard deviation and correlation.

Fig. 9. Performance of the methodology as a function of the price of electricity and discount factor.

The sensitivity analysis shows that, even if the estimations of correlation and standard deviation of the wind resource are exact, the performance of the methodology is not optimal (i.e., is not the highest point of the curve). This occurs because the expected error of the WRA is modeled with just two monomial scenario trees to reduce computational burden. This assumption is acceptable for this study because, as Table V showed, the overall accuracy of the methodology is acceptable and, according to Fig. 8, the performance of the methodology is near optimal. The sensitivity analysis also suggests that overestimating the standard deviation and underestimating the correlation of the wind resource have the greatest impact on the performance of the methodology. Both errors result in overestimating the uncertainty of the WRA, and can lead the ROs methodology to recommend investment decisions after the optimal time period. This would increase the costs of the WPPs and reduce the value of ROs. Regardless, the mean value of ROs remained positive within the scope of the sensitivity analysis. The power curve, electricity price, and discount factors directly affect the cash ows of the WPPs. These cash ows can become low due to low electricity prices and generation output (due to a cubic power curve), and high discount factors. In such cases, the differences between the NPV of immediate and a postponed investment decision is low, making investment decisions deferral more risky and decreasing the accuracy of the ROs methodology. The opposite occurs when the cash ows are high. The results do not show clearly the relation between the value of ROs and the different power curves, electricity price, and discount factors. Additional simulations using a single case study are performed to study these relations; see Fig. 9. Variations in the power curves are not included because they cause similar effects on the NPV of WPPs as electricity price variations. Fig. 9 shows that the value of ROs decreases when the price of electricity and the discount factor are either too low or high and, as the previous results showed, the accuracy of the methodology increases with high electricity prices and low discount factors. If the price of electricity is low, the WPP is more likely to be abandoned and an RO to delay the project to extend the WRA becomes less valuable. If the discount factor is low, the value of ROs decreases because the expected present value of the costs due to securing land and extending the WRA become higher. If

the electricity price or discount factor are too high, the expected benets lost due to delaying the project become more signicant, which decreases the expected value of ROs. VI. CONCLUSION This paper proposes an ROs methodology to incorporate WRA uncertainty to the planning and design process of WPPs. The idea is that project planners possess ROs to delay construction of WPPs with the objective of gathering additional data for the WRA in expectance that the wind forecast would become more accurate and allow the formulation of better designs for WPPs, and thus investments in more valuable projects. The proposed ROs methodology is meant to complement existing tools for the planning and design of WPPs. Existing WRA and WPP planning and design tools can be incorporated to the methodology in its error forecast stage and scenario trees assessment stage, respectively. The novelties of this work are that it presents an ROs methodology with the potential to increase the expected value of WPPs, and it extends ROs theory and WPP literature by studying exibility within the design of WPPs, and addressing the impact of WRA uncertainty in the planning and design process of WPPs. The implementation of the methodology is thoroughly explained and illustrated with a small case study. It is recognized that the performance of the methodology is case specic and affected by underlying assumptions. Therefore, the methodology is tested using a wide range of case studies with several considerations. The results show the impact of different assumptions on the accuracy of the methodology to identify optimal investment timing, and on the expected value of ROs. It is important to stress that the results presented in this work are based on a limited amount of wind sites, only one source of uncertainty, simulated data, and several assumptions to reduce computational burden. Further studies are needed to corroborate and generalize the main ndings of this work. Nevertheless, the results clearly illustrate the potential of the ROs methodology to enhance the value of WPPs and the circumstances that improve and weaken its performance. REFERENCES
[1] (2009) Wind Energy: The Facts European Wind Energy Association [Online]. Available: http://www.wind-energy-the-facts.org/, [Accessed: Apr. 21, 2010]

166

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 3, NO. 1, JANUARY 2012

[2] G. Kumbaroglu, R. Madlener, and M. Demirel, A real options evaluation model for the diffusion prospects of new renewable power generation technologies, Energy Econ., vol. 30, no. 4, pp. 18821908, 2008. [3] R. Prasad, R. Bansal, and M. Sauturaga, Some of the design and methodology considerations in wind resource assessment, IET Renew. Power Generation, vol. 3, no. 1, pp. 5364, 2009. [4] B. Bailey, Wind Resource Assessment Handbook. Albany, NY: AWS Scientic, 1997. [5] J. C. Mora, J. M. C. Barn, J. M. R. Santos, and M. B. Payn, An evolutive algorithm for wind farm optimal design, Neurocomputing, vol. 70, no. 1618, pp. 26512658, 2007. [6] C. N. Elkinton, J. F. Manwell, and J. G. McGowan, Algorithm for offshore wind farm layout optimization, Wind Eng., vol. 32, no. 1, pp. 6783, 2008. [7] S. Roy, Optimal planning of wind energy conversion systems over an energy scenario, IEEE Trans. Energy Convers., vol. 12, no. 3, pp. 248254, Sep. 1997. [8] S. Roy, Market constrained optimal planning for wind energy conversion systems over multiple installation sites, IEEE Power Eng. Rev., vol. 22, no. 1, p. 67, Jan. 2002. [9] J. S. Gonzalez, A. G. Rodriguez, J. C. Mora, M. B. Payan, and J. R. Santos, Overall design optimization of wind farms, Renew. Energy, vol. 36, no. 7, pp. 19731982, 2011. [10] J. S. Gonzalez, A. G. G. Rodriguez, J. C. Mora, J. R. Santos, and M. B. Payan, Optimization of wind farm turbines layout using an evolutive algorithm, Renew. Energy, vol. 35, no. 8, pp. 16711681, 2010. [11] M. Banzo and A. Ramos, Stochastic optimization model for electric power system planning of offshore wind farms, IEEE Trans. Power Syst., vol. 26, no. 3, pp. 13381348, Aug. 2011. [12] S.-D. Kwon, Uncertainty analysis of wind energy potential assessment, Appl. Energy, vol. 87, no. 3, pp. 856865, 2010. [13] M. A. Lackner, A. L. Rogers, and J. F. Manwell, Uncertainty analysis in MCP-based wind resource assessment and energy production estimation, J. Solar Energy Eng., vol. 130, no. 3, p. 031006, 2008. [14] A. K. Dixit and R. S. Pindyck, Investment Under Uncertainty. Princeton, NJ: Princeton Univ. Press, 1994. [15] S. C. Myers, Determinants of corporate borrowing, J. Fin. Econ., vol. 5, no. 2, pp. 147175, 1977. [16] K. L. Hastie, One businessmans view of capital budgeting, Fin. Manage., vol. 3, no. 4, pp. 3644, 1974. [17] L. Trigeorgis, Real Options: Managerial Flexibility and Strategy in Resource Allocation. Cambridge, MA: MIT Press, 1996. [18] F. Black and M. S. Scholes, The pricing of options and corporate liabilities, J. Political Econ., vol. 81, no. 3, pp. 63754, May/Jun. 1973. [19] J. C. Cox, S. A. Ross, and M. Rubinstein, Option pricing: A simplied approach, J. Fin. Econ., vol. 7, no. 3, pp. 229263, Sep. 1979. [20] T. E. Hoff, R. Margolis, and C. Herig, A Simple Method for Consumers to Address Uncertainty When Purchasing Photovoltaics Napa, CA, Sep. 2003 [Online]. Available: http://www.cleanpower.com/Research, cleanpower research [21] J. Sarkis and M. Tamarkin, Real options analysis for renewable energy technologies in a GHG emissions trading environment, in Emissions Trading, R. Antes, B. Hansjrgens, and P. Letmathe, Eds. New York: Springer, 2008, pp. 103119. [22] C.-H. Wang and K. Min, Electric power generation planning for interrelated projects: A real options approach, IEEE Trans. Eng. Manage., vol. 53, no. 2, pp. 312322, May 2006. [23] J. Munoz, J. Contreras, J. Caamano, and P. Correia, Risk assessment of wind power generation project investments based on real options, in Proc. IEEE PowerTech, Bucharest, Jun. 28Jul. 2, 2009, pp. 18. [24] A. Botterud, M. Ilic, and I. Wangensteen, Optimal investments in power generation under centralized and decentralized decision making, IEEE Trans. Power Syst., vol. 20, no. 1, pp. 254263, Feb. 2005. [25] P. Correia, P. Carvalho, L. Ferreira, J. Guedes, and J. Sousa, Power plant multistage investment under market uncertainty, Generation, Transmission Distribution, IET, vol. 2, no. 1, pp. 149157, Jan. 2008. [26] E. A. Martinez-Cesena and J. Mutale, Application of an advanced real options approach for renewable energy generation projects planning, Renew. Sustain. Energy Rev., vol. 15, no. 4, pp. 20872094, 2011. [27] T. Wang, Real Options in Projects and Systems Designs: Identication of Options and Solution for Path Dependency, Ph.D. dissertation, Massachusetts Institute of Technology, Cambridge, MA, 2005. [28] R. Neufville, Real options: Dealing with uncertainty in systems planning and design, Integrated Assessment, vol. 4, no. 1, pp. 2634, 2003.

[29] R. de Neufville, S. Scholtes, and T. Wang, Real options by spreadsheet: Parking garage case example, J. Infrastructure Syst., vol. 12, no. 2, pp. 107111, 2006. [30] N. Gil, Project safeguards: Operationalizing option-like strategic thinking in infrastructure development, IEEE Trans. Eng. Manage., vol. 56, no. 2, pp. 257270, May 2009. [31] T. Wang and R. Neufville, Building real options into physical systems with stochastic mixed-integer programming, in Proc. 8th Annu. Real Options Int. Conf., Montreal, Canada, 2004. [32] S. M. Ross, An Introduction to Mathematical Finance: Options and Other Topics. Cambridge, U.K.: Cambridge Univ. Press, 1999. [33] T. Wang and R. Neufville, Real options in projects, in Proc. 9th Annu. Real Options Int. Conf., Paris, France, 2005. [34] A. L. Rogers, J. W. Rogers, and J. F. Manwell, Comparison of the performance of four measure-correlate-predict algorithms, J. Wind Eng. Industrial Aerodynamics, vol. 93, no. 3, pp. 243264, 2005. [35] C. G. Justus, K. Mani, and A. S. Mikhail, Interannual and month-tomonth variations of wind speed, J. Appl. Meteorol., vol. 18, no. 7, pp. 913920, 1979. [36] G. M. Masters, Renewable and Efcient Electric Power Systems. Piscataway, NJ, Chichester: Wiley, 2004. [37] C. Kongnam, S. Nuchprayoon, S. Premrudeepreechacharn, and S. Uatrongjit, Decision analysis on generation capacity of a wind park, Renew. Sustain. Energy Rev., vol. 13, no. 8, pp. 21262133, 2009. [38] J. F. Manwell, J. G. McGowan, and A. L. Rogers, Wind Energy Explained: Theory, Design and Application. Chinchester: Wiley, 2002. [39] T. H. Cormen, C. E. Leiserson, R. L. Rivest, and C. Stein, Introduction to Algorithms, 2nd ed. New York: McGraw-Hill Science/Engineering/Math, Dec. 2003. [40] B. W. Raichle and W. R. Carson, Wind resource assessment of the southern appalachian ridges in the southeastern united states, Renew. Sustain. Energy Rev., vol. 13, no. 5, pp. 11041110, 2009. [41] M. Arendz, F. o. Egbrink, H. v. d. Berge, and R. Weijl, Catalogue. World Wide Wind Turbines Molenhoek, The Netherlands, 2007 [Online]. Available: http://www.Ref41urbines.com/en/, [Accessed: Mar. 18, 2010] [42] M. L. A. Fingersh and L. Hand, Wind Turbine Design Cost and Scaling National Renewable Energy Laboratory, West Covina, CA, Tech. Rep. NREL/TP-500-40566, 2006 [Online]. Available: http://www.nrel.gov/ wind/pdfs/40566.pdf [43] J. C. Helton and F. J. Davis, Latin hypercube sampling and the propagation of uncertainty in analyses of complex systems, Reliabil. Eng. Syst. Safety, vol. 81, no. 1, pp. 2369, 2003.

Eduardo Alejandro Martinez-Cesena (S06) received the B.Eng. degree from the Universidad Autonoma de Baja California, Baja California, Mexico, in 2005, and the M.Sc. degree from Instituto Tecnologico de Morelia, Michoacan, Mexico, in 2008, both in electrical engineering. He is currently working toward the Ph.D. degree in electrical and electronic engineering at the University of Manchester, Manchester, U.K. His research areas include the planning and design of generation systems based on renewable energies, especially wind farms, photovoltaic systems, and hydropower systems, as well as Real Options theory, power system economics, optimization techniques, and the smart grid.

Joseph Mutale (S97A00M03SM06) received the B.Eng. degree from the University of Zambia, and the M.Sc. and Ph.D. degrees from UMIST, all in electrical engineering. He is currently a Senior Lecturer in the School of Electrical and Electronic Engineering at the University of Manchester, Manchester, U.K. His research interests include power system economics, integration of new and renewable generation into power systems, as well as planning and operation of sustainable electric power systems for developing countries.