Collège doctoral T H E S E. pour obtenir le grade de Docteur de l Ecole des Mines de Paris Spécialité Energétique - PDF

Description
Collège doctoral N attribué par la bibliothèque T H E S E pour obtenir le grade de Docteur de l Ecole des Mines de Paris Spécialité Energétique présentée et soutenue publiquement par Pierre Pinson le 23

Please download to get full document.

View again

of 222
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Information
Category:

Slides

Publish on:

Views: 78 | Pages: 222

Extension: PDF | Download: 0

Share
Transcript
Collège doctoral N attribué par la bibliothèque T H E S E pour obtenir le grade de Docteur de l Ecole des Mines de Paris Spécialité Energétique présentée et soutenue publiquement par Pierre Pinson le 23 mars 2006 ESTIMATION OF THE UNCERTAINTY IN WIND POWER FORECASTING (ESTIMATION DE L INCERTITUDE DES PREDICTIONS DE PRODUCTION EOLIENNE) Directeur de thèse : Georges KARINIOTAKIS Jury : Pr. Arthouros ZERVOS (NTUA, Greece)...Président Pr. Henrik MADSEN (DTU, Denmark)...Rapporteur Pr. Vladimiro MIRANDA (INESC, Portugal)...Rapporteur Dr. Frédéric ATGER (Météo-France, France)... Examinateur Dr. Etienne BRIERE (EDF, France)... Examinateur Dr. Georges KARINIOTAKIS (ENSMP, France)... Examinateur . Forecasting is very difficult, especially if it s about the future... NielsBohr If a man will begin with certainties, then he shall end in doubts; but if he will content to begin with doubts he shall end in certainties... FrancisBacon . Jury: Pr. Arthouros Zervos, National Technical University of Athens, Greece (Président) Pr. Henrik Madsen, Technical University of Denmark, Denmark (Rapporteur) Pr. Vladimiro Miranda, Institute for Systems and Computer Engineering, Portugal (Rapporteur) Dr. Frééric Atger, Météo France, France (Examinateur) Dr. Etienne Brière, Electricité de France, France (Examinateur) Dr. Georges Kariniotakis, Ecole des Mines de Paris, France (Examinateur) ii Abstract WIND POWER experiences a tremendous development of its installed capacities in Europe. Though, the intermittence of wind generation causes difficulties in the management of power systems. Also, in the context of the deregulation of electricity markets, wind energy is penalized by its intermittent nature. It is recognized today that the forecasting of wind power for horizons up to 2/3-day ahead eases the integration of wind generation. Wind power forecasts are traditionally provided in the form of point predictions, which correspond to the most-likely power production for a given horizon. That sole information is not sufficient for developing optimal management or trading strategies. Therefore, we investigate on possible ways for estimating the uncertainty of wind power forecasts. The characteristics of the prediction uncertainty are described by a thorough study of the performance of some of the state-of-the-art approaches, and by underlining the influence of some variables e.g. level of predicted power on distributions of prediction errors. Then, a generic method for the estimation of prediction intervals is introduced. This statistical method is non-parametric and utilizes fuzzy logic concepts for integrating expertise on the prediction uncertainty characteristics. By estimating several prediction intervals at once, one obtains predictive distributions of wind power output. The proposed method is evaluated in terms of its reliability, sharpness and resolution. In parallel, we explore the potential use of ensemble predictions for skill forecasting. Wind power ensemble forecasts are obtained either by converting meteorological ensembles (from ECMWF and NCEP) to power or by applying a poor man s temporal approach. A proposal for the definition of prediction risk indices is given, reflecting the disagreement between ensemble members over a set of successive look-ahead times. Such prediction risk indices may comprise a more comprehensive signal on the expected level of uncertainty in an operational environment. A probabilistic relation between classes of risk indices and the level of forecast error is shown. In a final part, the trading application is considered for demonstrating the value of uncertainty estimation when predicting wind generation. It is explained how to integrate that uncertainty information in a decision-making process accounting for the sensitivity of end-users to regulation costs. The benefits of having a probabilistic view of wind power forecasting are clearly shown. iii ABSTRACT iv Préface I have started with the research works gathered in this thesis in October 2002, almost by coincidence. The path I was following was definitely not leading me towards pursuing Ph.D. studies. Though, series of fortunate events made that my resume ended up in the hands of Georges Kariniotakis, at the Centre for Energy and Processes of École des Mines de Paris, who proposed me to work with him in the field of wind power forecasting. I kindly thank Gilles Guérassimoff, François Neirac, Didier Mayer, and of course Georges for having given me that possibility. I must also thank all these fortunate events... The public defense of the thesis took place on March 23 rd, 2006, in Sophia Antipolis. Professional acknowledgements All the participants in the European project ANEMOS are fully acknowledged for a fruitful collaboration. More particularly, I would like to further acknowledge ESB (Electricity Supply Board of Ireland), Elsam and IDAE (Instituto para la Diversificación y Ahorro de la Energía) for providing the datasets (both power measures and HIRLAM meteorological predictions) that are used in the present thesis. In addition, I must thank the members of the jury, of which Pr. Zervos accepted to be the president, and of which Pr. Madsen and Pr. Miranda have accepted to be the censors. All the members of the jury have provided me with valuable comments for the present thesis, and comforted me in future directions to give to my research on wind power forecasting. Personal acknowledgements This research area is a perfect combination of several topics I have always dreamed to deal with, namely mathematical modeling, meteorological prediction, and wind energy. In addition, my aim was to carry out research in an international framework, and this has been v PRÉFACE possible mainly thanks to my participation in the Anemos project. It has been a pleasure for me to work and collaborate with all the participants in that project: some of them allowed me to use data that were essential for my research work, and some others took time for discussing (or maybe better say arguing...) my opinions on some topics related to this work. During these last years, I have had the chance to meet experts in wind energy, meteorology and mathematics. Most of these meetings have significantly influenced my view of forecasting and decision-making. For instance, I remember my meeting with Frédéric Atger from Météo France in June I am grateful that he spent time for reviewing my preliminary developments and for the discussion we had on his view of forecasting. Similarly, I must thank the WPPT team (i.e. Henrik Madsen, Henrik Aa. Nielsen and Torben S. Nielsen), at the Technical University of Denmark, for having allowed me to come and collaborate with them on some particular points of my work. This collaboration has considerably helped me in developing a broader view on wind power forecasting, from some basic to more challenging problems. Obviously, I am grateful to the various members of the jury, who spent time for reading the thesis, and giving me comments on my proposals, assumptions and developments. Finally, I have to thank Georges Kariniotakis for having given me part of his precious time in order to transmit his expertise in wind energy and distributed generation. I am sorry to say that these experts have not been the only people to provide me with valuable contribution for developing this research work. For instance, I have had the pleasure to be assisted by two master students who participated in specific points. And, introducing them to wind power forecasting obligedmetohaveaclearviewonmyobjectives and ways to meet them. Thank you Nils and Christophe. I hope you will be successful with your Ph.D. studies. In addition, all my collegues at École des Mines de Paris are to be acknowledged for their professional and kind support. Special thanks are due to Alain who was always available for IT issues, and also because he saved my laptop few weeks before the defense. Now is the more tricky part where I have to thank relatives, and people that I consider as relatives.first,thereisthatgroupof crazyguys ImetduringmyfirstyearinToulouse,when starting my studies at the Institut National des Sciences Appliquées. We are best friends since that time and I know they have had a great influence in my approach to life. Thanks for that. Then, I have to thank all my family for their support, whenever and whatever. More particularly, I think about my parents, who continuously encouraged my choices and were present in bad moments. I also need to thank my twin brother for so many things that I cannot mention them here... Last, and certainly not least, Pernille, mange tak for your patience and your affection... vi Contents Abstract Préface Contents Abbrevations, Notations and Mathematical Symbols ii vii x xi 1 Introduction Generalcontext Forecastingwindpower Estimating the uncertainty of wind power forecasts Purposeofthework Structureofthethesis State of the Art in Wind Power Forecasting Introduction Describingthebasisoftheproblem The intermittent nature of wind generation The various motivations for forecasting wind generation Themainaspectsoftheforecastingproblem Generic formulation of the wind power forecasting problem Thereferenceforecastingmethods Thephysicalapproaches Thephysicalmethodology Overviewofphysicalmethods vii CONTENTS 2.6 Thestatisticalapproaches Thestatisticalmethodology Overviewofstatisticalmethods Conclusions Characterizing the Uncertainty of Wind Power Predictions Introduction Defining and measuring forecast accuracy Thepredictionerror Evaluationframework Definition of an appropriate evaluation protocol Scopeofthestudy Predictionmethods Case-studies Evaluating the quality of state-of the-art point prediction methods Analysisbasedonerrormeasures Performanceagainstreferenceapproaches Analysisbasedonerrordistributions Highlighting the characteristics of the prediction uncertainty Contributions to the wind power prediction error Characteristicsofprediction errors Conclusions Estimation and Evaluation of Prediction Intervals of Wind Power Introduction Differenttypesofstatisticalintervals Basic parametric approaches for prediction interval estimation Developmentofadistribution-free approach Hypothesis and development of empirical-type methods Classificationofforecastconditions Thefuzzyinferencemodel Methods for combining error distributions Applicationtothewindpowerforecastingproblem Discussiononoperationalaspects A non-parametric framework for the evaluation of prediction intervals Requiredpropertiesforintervalforecasts Methods for the evaluation of prediction intervals Results Linear opinion pool vs. Adapted resampling Influence of the fuzzy mapping of the forecast conditions Influenceofthesamplesize Influence of the number of bootstrap replications Conclusions viii CONTENTS 5 Ensemble Predictions of Wind Power for Skill Forecasting Introduction Ensemblepredictions ofwindpower The meteorological ensemble predictions from ECMWF and NCEP Conversiontoensemblesofwindpower Poorman sensemblesofwindpower Ensemblesvs.spotforecasts The possibility to derive more accurate point predictions The ensembles ability to reflect the forecast uncertainty Skillforecastsbasedonwindpowerensembles Skill forecasting in the wind power prediction literature Definition of prediction risk indices On the relation between NPRI and energy imbalance Pointwise estimation of expected uncertainty Estimation of the uncertainty for a look-ahead period Conclusions The Value of Forecasting and the Benefits from Uncertainty Estimation Introduction Tradingwindgenerationinelectricitymarkets DescribingtheEuropeanelectricitymarkets Assumptionsforthepresentstudy Formulationoftheproblem Definition of advanced bidding strategies Pointpredictions asthebestbids Advanced bidding strategies based on probabilistic forecasts Evaluation of bidding strategies on a European electricity pool SpecificitiesoftheDutchelectricitymarket Resultsanddiscussion Conclusions General Conclusions Overallconclusions andcontribution Perspectives A List of Publications 181 B Résumé en français 183 B.1 Introduction B.2 Etat de l art de la prédiction éolienne B.3 Caractérisation de l incertitude de prédiction B.4 Estimation et évaluation d intervalles de prédiction B.5 Prédiction ensemblistes et indices de risque B.6 Valeurdes prédictions éoliennes et de l information sur leur incertitude ix CONTENTS B.7 Conclusions C Implementation of an Online Module 203 D Point Forecasting Methods - Evaluation Results 207 D.1 Contentdescription D.2 TunøKnob D.3 Klim D.4 Golagh D.5 Sotavento E Uncertainty Characteristics - Full Survey 221 E.1 Contentdescription E.2 TunøKnob E.3 Klim E.4 Golagh E.5 Sotavento Bibliography 235 x Abbreviations ABBREVIATIONS, NOTATIONS AND MATHEMATICAL SYMBOLS a.g.l. ADEME APX AWPPS CFD CO 2 EWEA Fuzzy-NN HIRLAM HIRPOM i.i.d. IPP ISET LOCALS LS-SVR MAE MOS MSE NETA NMAE NN NO x NPRI NRMSE NSDE NWP PC PCA PTU RA RMSE RUC SCADA SDE SO 2 SVM TenneT TSO WPPT Above ground level French Environment and Energy Management Agency Amsterdam Power exchange Armines Wind Power Prediction System Computational Fluid Dynamics Carbon Dioxide European Wind Energy Association Fuzzy-Neural Networks High Resolution Limited Area Model HIRlam POwer prediction Model Independant and identically distributed Independant Power Producer Institut für Solare Energieversorgungstechnik Local Circulation Assessment and Prediction System Least-Square Support Vector Regression Mean Absolute Error Model Ouput Statistics Mean Square Error New Electricity Trading Arrangement Normalized Mean Absolute Error Neural Network Oxides of Nitrogen Normalized Prediction Risk Index Normalized Root Mean Square Error Normalized Standard Deviation of the Errors Numerical Weather Prediction Probabilistic Choice Principal Component Analysis Program Time Unit Risk Analysis Root Mean Square Error Rapid Update Cycle Supervisory Control And Data Acquisition Standard Deviation of the Errors Sulfur Dioxide Support Vector Machine TSO for the Netherlands Transmission System Operator Wind Power Prediction Tool xi ABBREVATIONS, NOTATIONS AND MATHEMATICAL SYMBOLS Notations and Mathematical Symbols (1 α) Nominal coverage rate a (α) Empirical coverage of the (1 α)-confidence prediction intervals A r Rotor swept area exposed to the wind A w Weibull scale parameter A Fuzzy set bias Bias of a prediction model B Number of bootstrap replications c Forecast condition C Set of forecast conditions δ (α) Width of the (1 α)-confidence prediction intervals d Imbalance ɛ Normalized prediction error e Prediction error E Amount of wind energy F Distribution function γ Performance ratio associated to bidding strategies Γ Gamma function G Cumulative distribution function Î (α) Prediction interval with nominal coverage rate (1 α) I (α) Indicator variable for the (1 α)-confidence prediction intervals κ Kurtosis of a probability distribution k Prediction horizon (alternatively referred to as look-ahead time) k w Weibull shape parameter ˆL (α) Lower bound of the (1 α)-confidence prediction interval N (µ, σ 2 ) Normal distribution, with mean µ and variance σ 2 Ω Set of prediction errors π Prices in the spot/regulation markets p Wind power P(X) Probability for the occurence of the event X P n Wind farm nominal power µ Mean of a probability distribution m Membership function related to a fuzzy set ν Skewness of a probability distribution ρ Coefficient of correlation ρ air Air density r (α) Quantile with proportion α of a probability distribution R Revenue of a participant in an electricity market Coefficient of determination R 2 xii ABBREVIATIONS, NOTATIONS AND MATHEMATICAL SYMBOLS. σ S Sc θ t t r T u Û (α) x t ˆx t+k/t v w ζ z z 0 Standard deviation of a probability distribution Sample of prediction errors Scoring rule for probabilistic forecasts Wind direction Time Temporal resolution of forecast series Imbalance/regulation costs Wind speed Upper bound of the (1 α)-confidence prediction interval Measured value of the x-variable at time t Prediction of the value of the x-variable at time t for time t + k Influential variable Weight of a fuzzy rule Wind power penetration Height above ground level Roughness length xiii ABBREVATIONS, NOTATIONS AND MATHEMATICAL SYMBOLS xiv C H A PT E R1 Introduction 1.1 General context TODAY, wind farm installations in Europe exceed 40 GW. Motivated by the Kyoto Protocol, the European Commission has set the target of doubling the share of renewables in gross energy consumption from 6% in 1997 to 12% in 2010 [66]. This directive targets 22,1% indicative share of electricity produced from renewable energy sources in total Community electricity consumption by To achieve this share, installed wind power capacity in the Member States should reach 45 GW. The European Wind Energy Association (EWEA) has set its own target to 60 GW, which was revised upwards in 2003 to 75 GW [242]. Wind energy is considered as the fastest growing technology in the landscape of the alternative power generating sources. Moreover, it appears to be a clean and cost-effective energy source [91]. Certain countries, such as Germany, Denmark and Spain, have managed to perform large-scale integration of wind generation on land 1. Future major developments of wind power capacities are more likely to take place offshore. Higher and more regular wind speeds [189], availability of space that permits to install large wind farms, and less difficulties with local population acceptance, are the main advantages of going offshore to produce electricity. Important offshore projects are currently in progress with Horns Rev being a pioneer wind farm, in operation since end of December This wind farm supplies alone 2% of the whole electricity consumption of Denmark [212]. Several other ambitious offshore projects are under study in some European countries like United Kingdom and 1 Installed wind power capacities in these three countries alone represent more than 80% of the total capacity installed in Europe. 1 Estimation of the Uncertainty in Wind Power Forecasting Germany among others. Indeed, offshore wind energy could be sufficient to feed the local demand in countries like United Kingdom or Denmark [1]. France has a peculiar position in this context. Despite the fact that it has one of the best wind power potentials of Europe [14] (similar to the one of the United Kingdom), wind energy struggles to take its place in the French energetic landscape. To achieve its energetic independence, France has chosen three decades ago to invest in nuclear power, and is now a world leader in this field. Moreover, although it seems that the French population is in majority in favor of wind power, even in the areas with a lot of wind parks [236], antiwind lobbies are very active in communicating their disagreement with the installation of wind farms 2. However, perspectives drawn by the ADEME (French Environment and Energy Management Agency) are rather optimistic for the next years: forecasts of the installed capacities for wind generation (including offshore installations) reach 7 GW in 2010 and 28 GW in 2020 [37, 38]. To support this development, incentive feed-in tariffs were set in June In addition, studies about the advantages of geographically dispersed wind generation throughout the country [8] aim at showing how wind will behave in the current French power system and at providing guidelines concerning the future installations. Either onshore or offshore, such a large-scale integration of wind generation is expected to cause several difficulties in the management of a power system, wind being highly variable by nature. Wind generation is traditionally seen as fatal by utilities: a high level of reserves is often allocated to account for the intermittent profile of wind production, thus reducing the benefits from the use of wind energy. Typically, a wind farm capacity fac
Related Search
Similar documents
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks