IEEE, 9th International Conference on Environment and Electrical Engineering, Prague, Czech Republic, May 16 – 19, 2010 1

Wind Power Forecasting & Prediction Methods

A. M. Foley1, Member, IEEE, P.G. Leahy2, Member, IEEE and E.J. McKeogh3, Member IEEE

the costs associated with wind power integration. In the last

Abstract — Globally on-shore wind power has seen five years a number of studies and reports have been
considerable growth in all grid systems. In the coming decade off- commissioned and carried out by consultants and academics to
shore wind power is also expected to expand rapidly. Wind power investigate the costs [7 - 10]. Interestingly the costs associated
is variable and intermittent over various time scales because it is
with the integration of other renewable or non-renewable
weather dependent. Therefore wind power integration into
traditional grids needs additional power system and electricity sources of power have gained less attention. This is probably
market planning and management for system balancing. This due to the fact that wind power is relatively unpredictable.
extra system balancing means that there is additional system costs Traditionally grid systems are designed around fully
associated with wind power assimilation. Wind power forecasting ‘dispatchable’ power plant using measurable risk in terms of
and prediction methods are used by system operators to plan unit loss of load probability (LOLP), loss of load expectation
commitment, scheduling and dispatch and by electricity traders
(LOLE), energy not served (ENS), expected energy not served
and wind farm owners to maximize profit. Accurate wind power
forecasting and prediction has numerous challenges. This paper (EENS) and value of lost load (VOLL). The cost of backing-
presents a study of the existing and possible future methods used up wind power with peaking plant with reserve power
in wind power forecasting and prediction for both on-shore and continues to be a difficulty for system operators.
off-shore wind farms. The recent slow down in world economies may result in a
Keywords- Meteorology, Wind, Wind power forecasting, Wind fall in demand for electricity. This will place additional
power prediction
pressure on grid systems with base load plant cycling during
the night-time valley. Such a development may increase costs
I. INTRODUCTION to integrate wind power, as well as normal operational costs.

R apid growth in wind power has led to the need for

advanced wind power forecasting and prediction methods.
Accurate wind power forecasting and prediction reduces the
Wind power forecasting and prediction tools are therefore
invaluable because they enable better dispatch, scheduling and
unit commitment planning of thermal and hydro generators as
risk in uncertainty and allows for better grid and power system wind power ramps up and down erratically. They reduce the
integration. By and large wind forecasting and prediction is risk and uncertainty of wind power production for all
predominantly focused on the immediate short-term of seconds electricity market participants. This paper presents a high-level
to minutes to the short-term of hours to typically two days and review of existing short term wind power forecasting and
to the medium term of 2 to 7 days. Short-term wind power prediction methods and identifies future possible
patterns are discussed in Reference [1]. Little research is developments for both on-shore and off-shore wind farms.
available in relation to long term wind power forecasting and
II. FORECASTING & PREDICTIONS METHODS
prediction. Long-term wind patterns and electricity demand is
studied in Reference [2]. There are a number of studies which Wind power forecasting and prediction is generally
study long-term wind patterns, but these focus primarily on the described in terms of physical methods, traditional statistical
effects of climate change [3 - 6]. The timescale of the model is or ‘black box’ methods and more recently the so-called
related directly to the application of the forecasting and learning approaches, artificial intelligence or ‘grey box’
prediction. Very short-term forecasts (milliseconds to minutes) methods. Hybrid methods can involve some aspect of all three.
are usually associated with the active control of turbines at the References [11 – 14] describe wind forecasting and prediction
wind farm and grid frequency and voltage stability. Whereas methods.
short-term forecasts and predictions (hours to 2 days) are Numerical weather prediction (NWP) models usually form
power system planning (unit commitment and dispatch the basis of source data for most wind power forecasting and
scheduling). Medium-term forecasts and predictions are prediction methods. As noted in Reference [15] NWP predicts
identified for maintenance and outages planning of the wind the weather, not just the wind. NWP forecasts are not
farms, thermal generators and the grid system. specifically produced for the electricity industry. NWP holds
There has also been much debate and discussion in regard to best for time horizons greater than 4 hours. A simple method
to predict the wind is to use persistence [16].
Persistence is a ‘dumb’ measure which can be used to
1
A M Foley, Lecturer, Civil & Environmental Engineering Department, School of Engineering,
University College Cork, College Rd., Cork, Republic of Ireland, Phone + 353 (0)21 427 2285 Email: benchmark other methods. Persistence is usually better than
aoife.foley@ucc.ie
2
P Leahy, Lecturer in Wind Energy Engineering, School of Engineering, University College Cork,
NWP methods for short-term prediction horizons of up about 3
College Rd., Cork, Republic of Ireland, Phone + 353 (0)21 427 2285 Email: p.leahy@ucc.ie
3
to 6 hours, whereas the climatologic mean is better for
E McKeogh, Senior Lecturer, Civil & Environmental Engineering Department, School of
Engineering, University College Cork, College Rd., Cork, Republic of Ireland, Phone + 353 (0)21 427 prediction horizons longer than 15 hours [17]. Improvement
2285 Email: e.mckeogh@ucc.ie
techniques for NWP are described in Reference [18]. Most

Authorized licensed use limited to: NED UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on September 16,2023 at 07:32:49 UTC from IEEE Xplore. Restrictions apply.
 IEEE, 9th International Conference on Environment and Electrical Engineering, Prague, Czech Republic, May 16 – 19, 2010 2

models are multi-step and provide look-ahead times for Computational fluid dynamics (CFD) is also used as an
numerous horizons but the bulk of these tools only produce a alternative method to adjust the local conditions for the
single expected value for each forecast timescale and are physical terrain. Some physical methods are described in
referred to as deterministic, spot or point forecasts. Hence their References [20 and 21].
use for stochastic optimization and risk assessment is limited
B. Statistical Methods
[19]. References [24 and 25] describe the state of the art in
wind power forecasting and prediction methods. In the statistical approach vast amount of data is analyzed
A number of wind power software models have been and the meteorological conditions are ignored. Hence the
developed internationally. Table 1 presents a non-exhaustive methods used are described as ‘blackbox’. Generally a
list of models. statistical relationship is developed between the weather
forecast or prediction and the potential power output from the
TABLE 1 wind farm [22]. Statistical techniques used include
SOME WIND POWER FORECASTING & PREDICTION MODELS autoregressive (AR), moving average (MA), autoregressive
Model Developer Method Geographical
Name Locations moving average model, (ARMA) and auto regressive
Prediktor L. Landberg at Physical Spain, Denmark, integrated moving average model (ARIMA) [23]. Learning
Risø Republic of approach methods include artificial neural networks (ANN),
Ireland,
Northern fuzzy logic, support vector machine (SVM) and a combination
Ireland, France, of a two or more learning approach methods. More recently
Germany, USA, they have been referred to as ‘grey box’ methods. They are
Scotland and
Japan
called learning approaches because they learn from the
WPPT IMM & DTU Statistical Denmark, relationship between the predicted wind and forecasted power
Australia, output from the historical time series [24].
Canada, Ireland
Holland, Sweden
and Greece III. ACCURACY OF MODELS
Zephyr Risø & IMM Hybrid Australia, Models are assessed and compared using mean error (bias),
Previento Oldenburg Hybrid Germany,
University Northern Ireland mean absolute error (MAE), mean square error (MSE), root
e Wind TM True Wind Inc. Hybrid USA mean square error (RMSE), histograms of the frequency
Sipreólico University Statistical Spain distribution of the error, the correlation coefficient, R, mean
Carlos III & absolute percentage error (MAPE) and the coefficient of
Red Eléctrica
de Espaňa determination, R2 [26].
WPMS ISET, Statistical Germany It is recommended that three measures are taken to reduce
Germany forecast and prediction errors. These are using a combination
WEPROG J. Jørgensen Hybrid Ireland,
& Denmark and
of different NWP models, shorter forecast horizons leads to
C. Möhrlen Germany lower forecast errors and increasing the spatial spread over a
at UCC larger geographical area. If many wind farms are grouped
GH Garrad Hassan Statistical Greece, Great
together for wind power forecasting and prediction purposes
Forecaster Britain and USA
LocalPred M. Perez at Hybrid Spain and the error decreases. An example of this is in Germany where
& RegioPred CENER, Ireland the forecast error for the aggregated wind power stays below
formerly 2.5% when the three control zones of E.ON, VET and RWE
CIEMET
Alea Wind Aleasoft at Statistical Spain are grouped together [27].
UPC, Spain In the USA an MAE of 10 to 15% for day-ahead modeling
Scirocco Aeolis Hybrid Netherlands, of the name plate capacity of the wind farm has been
Forecasting Germany and
Services, Spain
experienced. If the model is rerun a few hours ahead on the
Netherlands same day the MAE range is typically 5% of the name plate
SOWIE Eurowind Physical Germany, capacity of the wind farm [28]. The Danish system operator
GmbH, Austria and has had similar results [29]. The RMSE is usually 10% of
Germany Switzerland
installed capacity for most models [21].
In Ireland the system operators (EirGrid and SONI) have a
A. Physical Methods target accuracy of 6 – 8% [30]. The operators have quoted
In the physical approach a detailed description of the lower individual wind farm accuracy in the range of 10 – 20%. As
atmosphere is used. Physical wind power forecasting involves Fig. 1 illustrates some forecasts on the Irish system have had
transforming the wind speed provided by the local errors [31].
meteorological service or NWP model to the wind turbines at
the wind farm, by extrapolating this data up or down to the
correct hub height, taking account of the physical terrain and
then using the power curve for the specific turbine or the
logarithmic power law to convert to wind power.

Authorized licensed use limited to: NED UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on September 16,2023 at 07:32:49 UTC from IEEE Xplore. Restrictions apply.
 IEEE, 9th International Conference on Environment and Electrical Engineering, Prague, Czech Republic, May 16 – 19, 2010 3

null hypothesis is rejected [37].

If the error in wind power forecasting and prediction is
reduced then electricity markets can trade with more certainty.
Contracting errors as a function of time can be as high as 39%
for a forecasting lead time of 4 hours [34]. A simplified model
to quantify the potential economic benefit of short-term wind
forecast is presented in Reference [35].
The research indicates that wind power forecasting and
prediction is a very valuable tool for wind power integration.
Wind power will become more attractive to system operators
as wind power prediction and forecast model accuracy
improves and as models become easier to use. Furthermore,
more efficient thermal plant and a more diverse generation
portfolio mix, which include fast response plant, hydro plant
and energy storage plant (such as pumped hydro, compressed
air) also make wind power integration less operationally
intensive for system operators’. Finally, when cycling of
Fig. 1 Snapshot of Actual (blue) & Forecasted (red) Wind Power Generation
5th February 2010 [31] nuclear plant is achieved in the market place and when smart
grid technology and intelligent load management techniques
IV. CURRENT & FUTURE DEVELOPMENTS (such as controlled hot water heating, controlled space heating
and electric vehicle charging) are deployed, wind power will
There is currently very little published data on off-shore
become a far simpler energy source to manage. The grid
wind power prediction and forecasting. There are ambitious
plans to develop large off-shore wind farms (e.g. Horns Rev, system, other thermal generators and the management of the
Denmark, Arklow Bank, Ireland and Hornsea, UK). Reference power system are circa 70 years old. Wind power has really
[32] discusses some of the issues associated with off-shore just developed to the current levels in just the last 15 years.
wind farm forecasting and prediction. It has a number of
difficulties, including: V. CONCLUSION
• Current forecasting and prediction models are designed In conclusion this paper presented a high-level review of
for on-shore environment and still have errors, existing short term wind power forecasting and prediction
• Resource assessment is difficult due to completely methods and identifies future possible developments for both
different conditions, off-shore is vast flat and smooth on-shore and off-shore wind farms.
(with a variable roughness) and so weather fronts are felt
more acutely than on land. Therefore thermal effects, VI. ACKNOWLEDGMENTS
wake affects and coastal land mass affects are amplified.
• poor availability of meteorological data to validate NWP The authors acknowledge EirGrid, the transmission system
outputs for these offshore locations. operator in the Republic of Ireland for supplying system data.
Current indications are that off-shore wind power forecasting
and prediction methods involve adapting existing models and VII. REFERENCES
using CFD to adjust for the maritime conditions. References [1] L. Landberg, Short-term prediction of the power production from wind
[25 and 35] detail the most up-to-date research available for farms, Journal of Wind Engineering and Industrial Aerodynamics, 1999
[2] G. Sinden, Characteristics of the UK wind resource: Long-term patterns
off-shore wind power forecasting and prediction. and relationships to electricity demand, Energy Policy, 2005
Development activities are on-going in reducing error in [3] S.C. Pryor and R.J. Barthelmie, Climate change on wind energy: A
wind power forecasting and prediction, improving regionalized review, Renewable and Sustainable Energy Reviews, 2010
wind power forecasting and prediction for on-shore wind [4] D.J. Sailor, M. Smith and M. Hart, Climate change implications for wind
power resources in the Northwest United States, Renewable Energy,
farms and deriving methods for wind power forecasting and 2008
prediction for off-shore wind farms. It is possible that the use [5] P.B. Breslow and D.J. Sailor, Vulnerability of wind power resources to
of ensemble and combined weather prediction methods climate change in the continental United States, Renewable Energy,
together may enhance wind power forecasting [22]. 2002
[6] I. Bogardi and I. Matyasovszky, Estimating Daily Wind Speed Under
Reference [36] presents a new tool called the WILMAR Climate Change, Solar Energy, 1996
and ANEMOS scheduling MeThodology (WALT). The [7] H. Holttinen, B. Lemström, P. Meibom, H. Bindner, A. Orths, F. Van
purpose of this model is to reduce the number of thermal 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,
generators on stand-by or in reserve using the probability of L.Söder, G. Strbac, A. Shakoor, J.C. Smith, P. Parsons, M. Milligan, Y.
generation outages and load shedding are system reliability Wan, Design and Operation of Power Systems with Large Amounts of
criteria instead of generation adequacy based solely on Wind Power, State of the art report, VTT Working Papers 82, Espoo,
generation outage. The wind and load forecast errors are Finland, 2007
[8] ILEX and G. Strbac, Quantifying the System Costs of additional
modeled using a Gaussian stochastic variable approach. Renewable in 2020, ILEX for DTI, London, 2002
However, another study recently found that if the [9] H. Shiu, M. Milligan, B. Kirby and K. Jackson, California Energy
Kolmogorov-Smirnov test is applied to the error data then the Commission, California Renewables Portfolio Standard Renewable

Authorized licensed use limited to: NED UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on September 16,2023 at 07:32:49 UTC from IEEE Xplore. Restrictions apply.
 IEEE, 9th International Conference on Environment and Electrical Engineering, Prague, Czech Republic, May 16 – 19, 2010 4

Generation Integration Cost Analysis, PIER Public Interest Energy Systems, IEE Proceedings of Generation, Transmission and Distribution,
Research Program, 2006 1994
[10] James Cox, Impact of intermittency: how wind variability could change [33] G. Kariniotakis, P. Pinson, N. Sibert, G. Giebel and R. Barthelemie,
the shape of the British and Irish electricity markets. Summary report, Proceedings of the 2004 SeaTechWeek, 20th – 21st October, 2004
Poyry Energy (Oxford) Ltd., July, 2009 [34] G.N. Bathhurst and G. Strbac, Value of Combining Energy Storage and
[11] B. Fox, D. Flynn, L. Bryans, N. Jenkins, M. O’Malley, R. Watson and D. Wind in Short-term Energy and Balancing Markets, Electric Power
Milborrow, Wind Power Integration: Connection and System Operation Systems Research, 2003
Aspects, The Institution of Engineering and Technology (IET), 2007 [35] R.J. Barthelmie, F. Murray, S.C. Pryor, The economic benefit of short-
[12] T. Burton, D. Sharpe, N. Jenkins and E. Bossanyi, Wind Energy term forecasting for wind energy in the UK electricity market, Energy
Handbook, John Wiley & Sons, ISBN 0471489972, 2001 Policy, 2008
[13] T. Ackermann, Wind Power in Power Systems, John Wiley & Sons, [36] A.F. Gubina, A. Keane, P. Meibom, J. O’Sullivan, O. Goulding, T.
ISBN 13:9780470855089, 2005 McCartan and M. O’Malley, New Tool for Integration of Wind Power
[14] G. Boyle, Renewable Electricity and the Grid – The Challenge of Forecasting into System Operation, IEEE Bucharest Power Tech
Variability, Earthscan, ISBN 13:9781844074181, 2007 Conference, Bucharest, Romania, June 28th – July 2nd, 2009
[15] S.J. Lang and E.J. McKeogh, Forecasting Wind Generation, Uncertainty [37] V. Miranda, R. Bessa, J. Gama, G. Conzelmann and A. Botterud, New
and Reserve Requirement on the Irish Power System using an Ensemble Concepts in Wind Power Forecasting Models, WindPower 2009
Prediction System, Wind Engineering, November 2009 Conference and Exhibition Centre, Chicago Illinois, May 4th – 7th, 2009
[16] Y.-K. Wu and J.-S. Hong, A Literature Review of Wind Forecasting in
the World, IEEE Power Tech, Bucharest, June 28th – July 2nd, 2009
[17] G. Giebel, R. Brownsword and G. Kariniotakis, State-of-the-Art in
VIII. BIOGRAPHIES
Short-Term Prediction of Wind Power – A Literature Review, Version Aoife M. Foley (b 1973) is a lecturer in the Civil & Environmental
1.1, Project ANEMOS, 2003 Department at the School of Engineering at (University
[18] T.S. Nielsen, H. Madsen, H. Aa. Nielsen, P. Pinson, G. Kariniotakis, N. College Cork) UCC. She is currently finalizing her
Siebert, I. Marti, M. Lange, U. Focken, L. von Bremen, P. Louka and G. doctoral research on wind power using energy storage.
Galanis, Short-term Wind Power Forecasting Using Advanced Statistical She graduated from UCC with a BE (Hons) in Civil &
Methods, European Wind Energy Conference & Exhibition, EWEC Environmental Engineering (1996). She also has an MSc
2006, Athens, Greece, 2006 (Hons) in Transportation Engineering (1999) from Trinity
[19] J. Juban, N. Siebert and G.N. Kariniotakis, Probabilistic Short-term College Dublin (TCD). She was also awarded an EPA
Wind Power Forecasting for the Optimal Management of Wind CCRP Fellowship (2009). She is a Chartered Engineer
Generation, IEEE Power Tech 2007, 1st – 5th July, Lausanne, with Engineers Ireland (EI) (2001) and is a Member of the
Switzerland, 2007 IEEE (2008).
[20] B.H. Chowdhury, Analysis of Wind Power Integration with Power
System Planning, Proceedings of the 25th Intersociety, Energy Dr Paul G. Leahy (b 1972) is the Science Foundation Ireland (SFI) Lecturer
Conversion Engineering Conference, Honolulu, Hawaii, USA, 1990 in Wind Energy Engineering at UCC. Paul graduated with
[21] G. Giebel, L. Landberg, G. Kariniotakis and R. Brownsword, State-of- first class honours in Electrical & Electronic Engineering
the-Art on Methods and Software Tools for Short-Term Prediction of from UCC (1995). He completed his PhD from UCC
Wind Energy, European Wind Energy Conference & Exhibition, EWEC (2001) in plasma diagnostics at the UKAEA-Euratom
2003, Madrid, Spain, 2003 Fusion Association in Culham, UK. Paul also has
[22] M. Lange and U. Focken, New Developments in Wind Energy postdoctoral experience in land-atmosphere interactions
Forecasting, IEEE Power & Energy Society General Meeting, and climatic extremes with Professor Gerard Kiely in
Conversion and Delivery of Electrical Energy in the 21st Century, HYDROMET at UCC. He is a Member of the IEEE
Pittsburg, Pennsylvania, USA, 20th – 24th July, 2008 (2009).
[23] M. Lei, L. Shiyan, J. Chuanwen, L. Hongling and Z. Yan, A Review on
the Forecasting of Wind Speed and Generated Power, Renewable Dr Eamon J. McKeogh (b 1953) is a Senior Lecturer in the Department of
Sustainable Energy Reviews, 2009 Civil & Environmental Engineering at UCC and lead PI
[24] B. Ernst, B. Oakleaf, M. L. Ahlstrom, M. Lange, C. Möhrlen, B. Lange, in Energy at the ERI. He graduated with a BScEng (1974)
U. Focken and K. Rohrig, Predicting the Wind, IEEE Power & Energy and a PhD (1978) from Queens University Belfast (QUB)
Magazine, November/December 2007 and an MBA (1986) from UCC. He provided input to the
[25] G. Kariniotakis, I.H.-P. Waldl, I. Marti, G. Giebel, T.S. Nielsen, J. EU Green Paper on Renewable Energy Sources, a
Tambke, J. Usaola, F. Dieriech, A. Bocquet and S. Virlot, Next consultation document used to prepare the EU White
Generation Forecasting Tools in the Optimal Management of Wind Paper. He is a Chartered Engineer with the Institution of
Generation, IEEE 9th International Conference on Probabilistic Methods Engineers of Ireland (1998). He is a Member of the IEEE
Applied to Power Systems, KTH, Stockholm, June 11th – 15th, 2006 (2009).
[26] A. Costa, A. Crespo, J. Navarro, G. Lizcano, H. Madsen and E. Feitosa,
A Review on the Young History of the Wind Power Short-term
Prediction, Renewable & Sustainable Energy Reviews, 2008
[27] J.C. Smith, R. Thresher, R. Zavadil, E. DeMeo, R. Piwko, B. Ernst and
T. Ackermann, A Mighty Wind – Integrating Wind Energy into the
Electric Power System Is Already Generating Excitement, IEEE Power
& Energy Magazine, March/April, 2009
[28] M.L. Ahlstrom and R.M. Zavadil, The Role of Wind Forecasting in Grid
Operations & Reliability, IEEE PES Transmission and Distribution
Conference & Exhibition: Asia and Pacific, Dalian, China, 2005
[29] T. Ackermann, G. Ancell, L.D. Borup, P.B. Eriksen, B. Ernst, F.
Groome, M. Lange, C. Möhrlen, A.G. Orths, J. O’Sullivan and M. de la
Torre, Where the Wind Blows, IEEE Power & Energy Magazine,
November/December, 2009
[30] P. O’Donnell and D. Garrett, EirGrid and SONI Scheduling & Dispatch
Workshop, 2008
[31] EirGrid, wind generation data , available at:
http://www.eirgrid.com/operations/systemperformancedata/windgenerati
on/
[32] S.J. Watson, L. Landberg and J.A. Halliday, Application of Wind Speed
Forecasting to the Integration of Wind Energy into Large Scale Power

Authorized licensed use limited to: NED UNIV OF ENGINEERING AND TECHNOLOGY. Downloaded on September 16,2023 at 07:32:49 UTC from IEEE Xplore. Restrictions apply.