Energy Procedia 12 (2011) 770 – 778

1876-6102 © 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of University of Electronic Science and Technology of China (UESTC).doi:10.1016/j.egypro.2011.10.103

Available online at www.sciencedirect.com

Energy Procedia

Energy Procedia 00 (2011) 000–000

www.elsevier.com/locate/procedia

ICSGCE 2011: 27–30 September 2011, Chengdu, China

A Review of Wind Power Forecasting Models

Xiaochen Wang a, b*, Peng Guoc, Xiaobin Huanga, b

a School of Control Science and Engineering, North China Electric Power University, Beijing, Chinab Beijing Hua Dian Tian Ren Power Control Technique Co. Ltd., Beijing, China

c School of Control Science and Engineering, North China Electric Power University, Beijing, China..

Abstract

Rapid growth in wind power, as well as increase on wind generation, requires serious research in various fields. Because wind power is weather dependent, it is variable and intermittent over various time-scales. Thus accurate forecasting of wind power is recognized as a major contribution for reliable large-scale wind power integration. Wind power forecasting methods can be used to plan unit commitment, scheduling and dispatch by system operators, and maximize profit by electricity traders. In addition, a number of wind power models have been developed internationally, such as WPMS, WPPT, Prediktor, ARMINES, Previento, WPFS Ver1.0 etc. This paper provides a review on comparative analysis on the foremost forecasting models, associated with wind speed and power, based on physical methods, statistical methods, hybrid methods over different time-scales. Furthermore, this paper gives emphasis on the accuracy of these models and the source of major errors, thus problems and challenges associated with wind power prediction are presented.

Keywords: Wind power forecasting; models for wind prediction; physical approaches; statistical approaches

1. Introduction

Wind energy is one of the RES characterized by the lowest cost of electricity production and the largest resource available. Therefore, a number of countries are beginning to recognize that wind power provides a significant opportunity for future power generation. As a result, the installed wind capacity grows more than 30% each year. And according to wind energy and green peace organization plan, 12%of all electricity generation should be achieved through wind power by 2020, with about 30GW [1].

As wind integration grows dramatically, the requirements for solving various problems, which includeeffective market design, Electricity Market Clearing, real-time grid operations, ancillary service

* Corresponding author. Tel.: +86-15210800227E-mail address: xiaochen_angel@126.com

Open access under CC BY-NC-ND license.

© 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of University of Electronic Science and Technology of China

Open access under CC BY-NC-ND license.

(UESTC)

Xiaochen Wang et al. / Energy Procedia 12 (2011) 770 – 778 7712 Xiaochen Wang et al. / Energy Procedia 00 (2011) 000–000

requirements and costs, competitive power quality, power system stability and reliability, transmission capacity upgrades and standards of interconnection, become more challenging [2]-[4].

However, improved wind forecasting can be considered as one of the most efficient way to overcomemany of these problems. Forecasting tools can enhance the position of wind by dealing with the intermittence nature of wind. Although wind energy may not be dispatched presently, the cost impacts of wind can be reduced to a large extent if the wind energy can be scheduled making use of accurate wind forecasting. Hence, the improvement of the performance of wind power and wind speed forecasting tool has significant economic and technical impact as well on the system by increasing wind power penetration.

Many studies have been devoted to the improvements of wind forecasting techniques by a number ofinstitutes and organizations with wide experience in the field. A number of models have been developed and launched on wind farms in world-wide locations, such as WPMS, WPPT, Prediktor, ARMINES,Previento, etc.. These models based on physical approaches, Statistical approaches and hybrid approaches are installed by end-users for on-line operation and evaluation at a local, regional and national scale for several years. To study the experience of elder generation as well as the precept of the mature patterns is an essential comprehension of state of the art of prediction and energy management systems.

This paper presents a detailed review on existing tools used in wind speed and wind power predictionover time-scales, and further identifies possible developments in the future. Generally speaking, wind forecasting is predominantly focused on the immediate-short-term of minutes to the short-term of hours to typically up to 1 day and to the long-term of up 2 days. As an example of immediate-short-term models, WPMS provides wind power prediction for over 95% territories of the whole Germany now. Immediate-short-term wind forecasting models are discussed in Reference [5]. In addition, several models, which are generally based on high precision numerical weather prediction (NWP), have been developed for short-term wind forecasting, such as Predictor, Zephyr, AWPPS, Ewind [6]-[10]. Previento, which is based on an hybrid approaches can provide wind forecasting for a time horizon up to 48 hours. More studies on long-term forecasting models are involved in reference [11]-[14].

2. Classification of Wind Forecasting

Various methods classified according to time-scales or methodology, are available for wind forecasting.

Actually, classification of wind forecasting approaches in terms of time-scale is unclear. However, making reference to a great deal of studies in this field [15]-[16], wind forecasting can be separated based on the prediction horizon into three categories: Immediate-short-term (8hours-ahead) forecasting, Short-term (day-ahead) forecasting, Long-term (multiple-days-ahead) forecasting.

Moreover, as shown in Table 1, applications of specific time horizon in electricity systems are consequential different.

Wind forecasting schemes can also be classified based on their methodology into two categories: Physical approach (deterministic approach)

Physical method or deterministic method is based on lower atmosphere or numerical weather prediction (NWP) using weather forecast data like temperature, pressure, surface roughness and obstacles.In general, wind speed obtained from the local meteorological service and transformed to the wind turbines at the wind farm is converted to wind power [17]. Statistical approach

Statistical method is based on vast amount of historical data without considering meteorological

772 Xiaochen Wang et al. / Energy Procedia 12 (2011) 770 – 778Xiaochen Wang et al. / Energy Procedia 00 (2011) 000–000 3

conditions. It usually involved artificial intelligence (neural networks, neuro-fuzzy networks) and time series analysis approaches [18]-[19]. Hybrid approach

Hybrid method，which combines physical methods and statistical methods， particularly usesweather forecasts and time series analysis.

Table 1. Time-horizon classification for wind forecasting

Time-scale Range ApplicationsImmediate-short-term 8hours-ahead - Real-time grid operations

- Regulation actionsShort-term Day-ahead - Economic load dispatch planning

- Load reasonable decisions- Operational security in electricity market

Long-term Multiple-days-ahead - Maintenance planning - Operation management- Optimal operating cost

3. Wind Forecasting Models

A general overview of wind forecasting models is presented in Table II. This section is divided intothree parts based on the time-scales, and for each of them, some typical models are described in detail with consideration, input data, methods, characteristics ect..

3.1. Immediate-short-term forecasting

Models for immediate-short-term forecasting are generally based on statistical approaches, especially ANN, because of the time-consuming of operating the NWP [20]-[21].

As one prominent example for immediate-short-term wind forecasting, WPMS have been modified for operation in the ICT environments of different grid operators and carriers of large wind parks [22].

WPMS used artificial neural networks (ANN) which are trained with a large deal of historical data in wind farm. Input data, the output data measured in wind farm and the meteorological parameters forecasted were converted by a preprocessor into XML-format and transferred to the programme core, consisting of prediction modules and transformation modules. The core IT makes the following calculations in turn:

Table 2. Wind speed and power forecasting models

Model Developer Time horizon ApproachWPMS ISET,Germany Immediate-short-term StatisticalWPPT IMM & DTU Short-term StatisticalPrediktor Risø Short-term PhysicalZephyr Risø& IMM Short-term Statistical&physicalWPFS Ver1.0 Chinese Electric Power Science Institute Short-term Statistical&physical

ANEMOS, 26 partners from 7 countries Immediate-short-term short-term Statistical&physicalARMINES(AWPPS)

European Commission Immediate-short-term Short-term Statistical&physical

Ewind AWS Truewind Short-term Statistical&physicalSipreolico University Carlos III & Red Eléctrica de Espaňa Short-term Statistical

Previento Oldenburg University Long-term Statistical&physicalLocalPred& RegioPred

CENER Long-term Statistical&physical

WEPROG UCC Long-term Statistical&physical

Xiaochen Wang et al. / Energy Procedia 12 (2011) 770 – 778 7734 Xiaochen Wang et al. / Energy Procedia 00 (2011) 000–000

Determination of current wind power feed-in for the entire control-rule-zone and for any grid region and partial regions.

Compilation of the next-day-ahead forecasts of wind power feed-in for the rule-control-zone and for grid regions based on forecasted meteorological parameters.

Calculation of immediate-short-term expected wind power feed-in for an interval of 1-8 hours for the control-rule-zone and partial regions, based on meteorological parameters and power data measured.

3.2. Short term forecasting

Several tools have been developed for wind power forecasting for short-term-forecasting, such as WPPT, Predictor, Zephyr, Ewind, WPFS Ver1.0, AWPPS. These models have been implemented by anumber of case studies in Spain, Germany, Denmark, Ireland, Greece and France [23].

One classical model with extensive application for this time-scale is The Wind Power Prediction Tool (WPPT). It can be used for generating short-term (say up to 120 hours，actually 36 hours) predictions of the wind power production. The system is very flexible because it can give prediction values as a total covering not only a single wind farm (like the northern part of Jutland), but also a region (like the Horns Rev off-shore wind farm). The system also provides reliable estimates of the uncertainty of the tools, which is very important for optimal scheduling or trading. WPPT is based on advanced non-linear statistical models. The set of models includes a semi-parametric power curve model for wind farms taking into account both wind speed and direction, and dynamical forecasting models describing the dynamics of the wind power and any diurnal variation, etc.

The models are self-calibrating and self-adaptive. As a result, they automatically adjust parametersaccording to changes in the number of turbines and their characteristics, surroundings, the NWP modelsand non-explicit characteristics of models like roughness, dirty blades.

WPPT based on artificial intelligence can automatically calibrate to the observed situation [24]. In the minimal setup, the system requires online measurements of wind power. However, depending on the configuration, the following data is taking into account: Online measurements of wind power, Aggregated energy readings form all (or almost all) turbines in a region (for regional forecasting), Meteorological forecasts of wind speed and direction covering wind farms and regions, Other measurements or predictions available like local wind speed, stability, number of active

turbines.Another important tool is Prediktor developed by the meteorology research programme (MET).

However unlike WPPT, the basic idea behind Prediktor is to try to model as much as possible using physical models. The system provides the expected production of wind farms up to typically 48 hours every 6 hours. All it need is an on-line access to output from a NWP model. The basic steps are as follows: the overall weather patterns are predicted by a NWP model. Such a model can only predict the overall wind and get valid predictions at a specific location. And then these predictions are tailored as necessary. This tailor-making is done by the WAsP model, which models the local influences, including the effects of roughness, orography (ridges and hills), and obstacles together with the influence that the wind turbines have on each other (that is the park effects).

Since no model can simulate nature perfectly, two MOS (model output statistics) filters are used in Prediktor to correct short-comings. The wind power observed is used to adjust the parameters of these filters. The final output of the model is the expected production of the wind farm every 3 hours over the next 48 hours. Furthermore, Prediktor forecasts or will forecast in the near future for up to 50 wind farms in Ireland, Northern Ireland, Denmark, Germany, France and Spain.

AWPPS is well worth mentioning in this fraction, which provides short-term forecasts for the power

774 Xiaochen Wang et al. / Energy Procedia 12 (2011) 770 – 778Xiaochen Wang et al. / Energy Procedia 00 (2011) 000–000 5

output of onshore and offshore wind farms for the next 48/72 hours with a time step of 1 hour (updates every hour), for the next 4-6 hours with a time step of 10-15 min (updates every 10-15 min), and on-line uncertainty assessment for these forecasts.

The Power Prediction Module provides forecasts for the output of each considered wind farm. The core module is based on state-of-the-art adaptive fuzzy neural networks. This approach provides strong advantages compared to classical neural networks or other statistical or physical techniques. The module is enhanced with on-line adaptation capabilities for optimal performance [25]-[27].

The AWPPS is the only available tool that provides confidence intervals for wind power forecasts with a predefined level of confidence (i.e. 85%, 90%, and 95%). The intervals are produced based on an advanced approach appropriate to the wind prediction problem. The Prediction Risk Module permits to predict the uncertainty based on the expected weather stability for the next 24 hours. In addition， the on-line use of this module permits to develop appropriate strategies for maximising the value from the use of power forecasts. The AWPPS includes a module for forecasting regional or national wind power based on a sample of reference wind farms.

More and more attention has been pay to wind forecasting by Chinese researchers. And WPFS Ver1.0, the first wind power forecasting model developed by Chinese electric power science institute has launched in Heilongjiang Province recently.

The WPFS Ver1.0 system considered historical weather data, wind farm data and generationparameters is based on both physical method (fresh wind farms) and statistical method (existed wind farms). Fig. 1 shows the frame of WPFS Ver1.0, which includes five main modules.

Fig.1. Frame of WPFS Ver1.0 with Five basic modules communicated with each other. Prediction system database is the core module of this system, and like WPPT, AWPPS and other hybrid models, NWP is taking into consideration.

3.3. Long term forecasting

A few studies has been done on long-term wind forecasting approaches. And prediction tools for this time-scale are not abundance. Due to long ahead-forecasting duration, simple models can not meet therequirements anymore, so use of NWP or hybrid NWP models is considered.

Modern wind power prediction tools provide forecasts for a time horizon of up to several days inadvance and are typically based on NWP. In other words, all the information about the future evolution of the wind forecasting is provided by the NWP. The national weather services or private weather data providers offers a broad range of different NWP data which are suitable for wind speed and wind power forecasting. It is a trend of using NWP for long-term forecast in the future[28].

Previento is similar to Prediktor, but uses more stringent physical downscaling and specializedupscaling. It provides a reliable prediction of the expected wind power for any sites and regions in Germany, Europe and anywhere around the world up to 10 days in advance and with a time resolution of up to 15 minutes. The wind power prediction rests on the optimal combination of different weather models, while taking into account the local conditions of the wind farm surroundings as well as the NWP.

User interface Predition system database EMS interface

NWP

Prediction program

Xiaochen Wang et al. / Energy Procedia 12 (2011) 770 – 778 7756 Xiaochen Wang et al. / Energy Procedia 00 (2011) 000–000

The Previento system is based on a physical approach with input from a large scale weather prediction model like Lokalmodell of the German Weather Service. It models the boundary layer with regard to roughness, orography and wake effects. Important for the calculation of the wind speed at hubheight is the daily variation of the thermal stratification of the atmosphere which is used to change the logarithmic profile. Using the specific power characteristic of the turbine, the expected power output for single sites is calculated. The total electricity produced by wind within a certain region is calculated on the basis of selected wind farms.

3.4. Hybrid forecasting

It is universal for a tool to complete predictive functions in each aspect. ANEMOS is a hybrid wind prediction tool considering various time horizons. In the frame of the project, emphasis is given on the development of integrating high-resolution meteorological forecasts and appropriate prediction models for the offshore [29]-[30]. For the offshore case, marine meteorology will be considered as well as information by satellite-radar images. Furthermore, in order to estimate the benefit of forecasting in a model of the Nord Pool electricity market, the WILMAR project supported by the European Commission has developed the market model for the simulation of wind power forecasting. The aim of this project is to develop accurate models that outperform considerably actual state-of-the-art, for onshore and offshore wind resource forecasting with advanced physical, statistical, and combined approaches.

The project is designed with nine work-modules which address the technical targets as below: Data collection and evaluation need, Estimation of on-line operation, Off-line evaluation of forecasting models, Offshore forecasting, Development of physical methods, Development of statistical methods, Development of ANEMOS forecasting platform, Platform installation for on-line operation, General dissemination and assessment.

This system will be installed by several utilities for on-line operation at onshore and offshore wind farms for national wind prediction. The applications are characterised by different terrains and climatessurrounding wind farms. The on-line operation by the utilities will permit to validate the models and to analyse how predictions can contribute to a competitive integration of wind energy in the developing electricity market.

4. Accuracy and the Source of Error

In general, accurate and reliable forecasting models of wind forecasting are recognized as a major contribution for increasing wind penetration. Customarily, models are assessed using mean error ,mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), histograms of the frequency distribution of the error, the correlation coefficient, R, mean absolute percentage error (MAPE) and the coefficient of determination, R2 [31].

It is proved that the error of the prediction result will enlarge with larger time horizon. However, some of the tools improved accuracy with increased input and considerations [31]. For example, as one of the most mature commercial model for immediate-short-term prediction, WPMS has shown excellentperformance with the RMSE of 7%-19%. AWPPS was successfully adapted and validated for more than 35 onshore and offshore wind farms in Denmark, Germany, Greece, Ireland, Portugal, Spain and UK

776 Xiaochen Wang et al. / Energy Procedia 12 (2011) 770 – 778Xiaochen Wang et al. / Energy Procedia 00 (2011) 000–000 7

situated in different terrain types (flat or complex). The performance for single wind farm forecasting ranges between 2-5% (of the nominal wind farm power) for one-hour ahead predictions and 10-15% for 48 hours ahead. The performance for regional forecasting is 8-10% for 24 hours ahead. On the Contrast,The MAE of TureWind system is 10.8% of the name plate capacity for one-day ahead forecasting, and 11.7% for up one-day ahead. In china ,the forecast RMSE of WPFS is 16-19% with single wind farm and 11.67% with a number of farms. However, it is not quite acceptable for Prediktor applied by NOAA, a United States organizes, with the MAE of 2.4m/s.

For the first time, 11 different models ran the same case, and were evaluated using the same criteria 6 test cases in Spain, Denmark, Ireland and Germany with the same NWP input. As a result, no single model was best in all cases, and forecasting accuracy gets worse in complex terrain.

Lots of factors can affect the predictive result in various aspects. To transform the obtained weather information into the power of a wind turbine is the key problem in

wind forecasting, whereas, almost 80% error can be explained by the errors of NWP. Fresh regime based stochastic models for the spatio- temporal corrections explains more than 20% of

the errors for a 2 hour prediction. Large prediction errors are due to phase errors in the MET forecasts used as input for the tool like

WPPT. Because running high resolution and large covered area both is numerically too expensive, only few

models are able to distinguish between sea and land and can adjust the resolution accordingly.These recommendable measures are taken to reduce prediction errors: using a combination of

different NWP patterns, like ANEMOS, shorter forecast horizons leads to lower forecast errors and increasing the spatial spread over a larger geographical region. It is expected to decrease the errors ifwind farms are grouped together for wind forecasting [32].

5. Conclusions

This paper provides a review on different tools with various techniques used for developing wind farmpower prediction considering different time scales. Several forecasting models, which have their own characteristics, were discussed. In addition, emphasize was given on accuracy of prediction models and the source of error. It is difficult to evaluate the performance of various models, as the existing applications were in different ways. No forecasting model of them can perfect any condition, however , they has served as effective tools to maximize the power captured thus increasing the reliability of wind power for wind farms. It is expected that these developed models can provide lots of experience andinspires to other researchers.

References

[1] Long-term and medium-term development of renewable energy, 2007, State council of the People's Republic of China[2] Y-K Wu, and J-S Hong, “A literature review of wind forecasting technology in the world,” IEEE Power Tech 2007,

Lausanne, pp. 504-509, 1-5 July 2007.[3] H. Lund, “Large-scale integration of wind power into different energy systems,” Energy, vol. 30, no. 13, pp. 2402-2412,

Oct. 2005.[4] P.B. Breslow and D.J. Sailor, Vulnerability of wind power resources to climate change in the continental United States,

Renewabl Energy,2002[5] H. Holttinen, B. Lemström, P. Meibom, H. Bindner, A. Orths, F. Van Hulle,C. Ensslin, A. Tiedemann, L. Hofmann, W.

Winter, A. Tuohy, M. O’Malley, P. Smith, J. Pierik, J.O. Tande, A. Estanqueiro, E. Gomez, L.Söder, G. Strbac, A. Shakoor, J.C. Smith, P. Parsons, M. Milligan, Y. Wan, Design and Operation of Power Systems with Large Amounts of Wind Power, State of the art report, VTT Working Papers 82, Espoo, Finland, 2007.

Xiaochen Wang et al. / Energy Procedia 12 (2011) 770 – 778 7778 Xiaochen Wang et al. / Energy Procedia 00 (2011) 000–000

[6] L. Landberg, “ Short-term prediction of local wind conditions, ” Journal of Wind Engineering and Industrial Aerodynamics, vol. 89 235-245,2001.

[7] Sa’nchez I., “Short-term prediction of wind energy production,” International Journal of Forecasting, vol. 22, pp.43-56, 2006

[8] C. Potter, M. Ringrose, and M. Negnevitsky, "Short-term wind forecasting techniques for power generation", Proceedings of the Australasian Universities Power Engineering Conference, Brisbane,Australia, 2004, C.D. ISBN: 1-864-99775-3.

[9] S. Salcedo-Sanz, A. M. Perez-Bellido, E. G. Ortiz-Garcia, A. Portilla- Figueras, L. Prieto, and D. Paredes, “Hybridizing the fifth generation mesoscale model with artificial neural networks for short-term wind speed prediction,” Renewable Energy, vol. 34, no. 6, pp. 1451-1457, June 2009.

[10] G. Kariniotakis, I. Marti, et al., "What Performance Can Be Expected by Short-term Wind Power Prediction Models Depending on Site Characteristics?", in CD-Rom proceedings of the European Wind Energy Conference EWEC 2004, London, UK, 22-25 Nov. 2004.

[11] T. G. Barbounis, J. B. Theocharis, M. C. Alexiadis, and P. S. Dokopoulos, “Long-term wind speed and power forecasting using local recurrent neural network models,” IEEE Trans. Energy Convers., vol. 21, no. 1, pp. 273- 284, March 2006.

[12] G. Sinden, Characteristics of the UK wind resource: Long-term patterns and relationships to electricity demand, Energy Policy, 2005

[13] S.C. Pryor and R.J. Barthelmie, Climate change on wind energy: A review, Renewable and Sustainable Energy Reviews, 2010

[14] D.J. Sailor, M. Smith and M. Hart, Climate change implications for wind power resources in the Northwest United States, Renewable Energy, 2008.

[15] L. Lazic, G. Pejanovic, and M. Zivkovic, “Wind forecasts for wind power generation using the Eta model,” Renewable Energy, vol. 35, no. 6, pp. 1236-1243, June 2010

[16] J. Tambke, et al., “Short-term Forecasting of Offshore Wind Farms Production – Developments of the Anemos Project”. In Proc. of the European Wind Energy Conference 2006, Athens, Greece, 27/2-2/3 2006.

[17] M. Lange, and U. Focken, “New developments in wind energy forecasting,” IEEE Power and Energy Society General Meeting 2008 - Conversion and Delivery of Electrical Energy in the 21st Century, pp. 1-8, 20-24 July 2008.

[18] G. Giebel, G. Kariniotakis, and R. Brownsword, “The State-of-the-Art in Short-Term Prediction of Wind Power - Aliterature Review,[Online]: Available: http: //www.anemos-project.eu/download//ANEMOS_D1.1_StateOfTheArt_v1.1.pdf.

[19] A. R. Garcia, and E. De-La-Torre-Vega, “A Statistical wind power forecasting system – A Mexican wind-farm case study,”European Wind Energy Conference & Exhibition – EWEC Parc Chanot, Marseille, France, March 2009.

[20] G. Sideratos and N. Hatziargyriou, “Using radial basis neural networks to estimate wind power production”, 1-4244-1298-6/07/$25.00 © 2007 IEEE.

[21] C. W. Potter, and M. Negnevitsky, “Very short-term wind forecasting for Tasmanian power generation,” IEEE Trans. Power Syst., vol. 21, no. 2, pp. 965- 972, May 2006.

[22] ISET: Wind Energy Report Germany 1999/2000 – Annual Evaluation of WMEP, Kassel, 2000.[23] B. Candy, S. J. English, and S. J. Keogh, “A Comparison of the impact of QuikScat and WindSat wind vector products on

met office analyses and forecasts,” IEEE Trans. Geosci. Remote Sens., vol. 47, no.6, pp. 1632-1640, June 2009.[24] E. Cadenas, and W. Rivera, “Short term wind speed forecasting in La Venta, Oaxaca, Mexico, using artificial neural

networks,” Renewable Energy, vol. 34, no. 1, pp. 274-278, Jan. 2009.[25] E. Cadenas, and W. Rivera, “Wind speed forecasting in the South Coast of Oaxaca, Mexico,” Renewable Energy, vol. 32,

no. 12, pp. 2116-2128, Oct. 2007.[26] E. Cadenas, O.A. Jaramillo, and W. Rivera, “Analysis and forecasting of wind velocity in chetumal, quintana roo, using the

single exponential smoothing method,” Renewable Energy, vol. 35, no. 5, pp. 925-930, May 2010.[27] YANG Xiu-Yuan and XIAO Yang, Wind Speed and Generated power forecasting in wind farm, Proceedings of the CSEE,

Vol 25, No. 11, pp. 1~5, Jun．2005.[28] R. G. Kavasseri, and K. Seetharaman, “Day-ahead wind speed forecasting using f-ARIMA models,” Renewable Energy,

vol. 34, no. 5, pp. 1388-1393, May 2009.[29] http: //anemos.cma.fr/.[30] Short-term Forecasting Using Advanced Physical Modelling-The Results of the Anemos Project. Results from mesoscale,

microscale and CFD modelling”, In Proc. of the European Wind Energy Conference 2006, Athens, Greece, 27/2-2/3 2006.[31] Enrique Romero and Daniel Toppo, “Comparing support vector machines and feed forward neural networks with similar

hidden layer weights”, IEEE Trans. on Neural Networks, Vol 18, no 3. May 2007, pp 959 – 963.

778 Xiaochen Wang et al. / Energy Procedia 12 (2011) 770 – 778Xiaochen Wang et al. / Energy Procedia 00 (2011) 000–000 9

[32] H. Madsen, G. Kariniotakis, H.Aa. Nielsen, T.S. Nielsen, P. Pinson,"Standardising the Performance Evaluation of Short-Term Wind Power Prediction Models", Wind Engineering, vol 29, no. 6, pp. 475-489,December,2005.