Modeling and Stochastic Learning for Forecasting in High Dimensions PDF
Preview Modeling and Stochastic Learning for Forecasting in High Dimensions
Lecture Notes in Statistics 217 Proceedings Anestis Antoniadis Jean-Michel Poggi Xavier Brossat E ditors Modeling and Stochastic Learning for Forecasting in High Dimensions Lecture Notes in Statistics 217 EditedbyP.Bickel,P.Diggle,S.E.Fienberg,U.Gather, I.Olkin,S.Zeger Moreinformationaboutthisseriesat http://www.springer.com/series/694 Anestis Antoniadis • Jean-Michel Poggi (cid:129) Xavier Brossat Editors Modeling and Stochastic Learning for Forecasting in High Dimensions 123 Editors AnestisAntoniadis Jean-MichelPoggi DepartmentofStatistics LaboratoiredeMathématiques UniversityJosephFourier UniversitéParis-Sud Grenoble,France OrsayCedex,France XavierBrossat ElectricitédeFranceR&D,OSIRIS ClamartCedex,France ISSN0930-0325 ISSN2197-7186 (electronic) LectureNotesinStatistics ISBN978-3-319-18731-0 ISBN978-3-319-18732-7 (eBook) DOI10.1007/978-3-319-18732-7 LibraryofCongressControlNumber:2015941890 SpringerChamHeidelbergNewYorkDordrechtLondon ©SpringerInternationalPublishingSwitzerland2015 Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartof thematerialisconcerned,specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation, broadcasting,reproductiononmicrofilmsorinanyotherphysicalway,andtransmissionorinformation storageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilarmethodology nowknownorhereafterdeveloped. Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublication doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant protectivelawsandregulationsandthereforefreeforgeneraluse. Thepublisher,theauthorsandtheeditorsaresafetoassumethattheadviceandinformationinthisbook arebelievedtobetrueandaccurateatthedateofpublication.Neitherthepublishernortheauthorsor theeditorsgiveawarranty,expressorimplied,withrespecttothematerialcontainedhereinorforany errorsoromissionsthatmayhavebeenmade. Printedonacid-freepaper SpringerInternationalPublishingAGSwitzerlandispartofSpringerScience+BusinessMedia (www.springer.com) Preface Forecasting and time series prediction have seen a great deal of development and attention over the last few decades, fostered by an impressive improvement in observational capabilities and measurement procedures. Time series prediction is a challenge in many fields. In finance, one forecasts stock exchange or stock market indices; data processing specialists forecast the flow of information on theirnetworks;producersof electricityforecastthe electric load,andhydrologists forecast river floods. Many methods designed for time series prediction and forecastingperformwell(dependingonthecomplexityoftheproblem)onarather short-termhorizonbutareratherpooronalonger-termone.Thisisduetothefact thatthesemethodsareusuallydesignedtooptimizetheperformanceatshort-term predictionusingrelativelylowdimensionalmodels.Forlong-termprediction,linear andnonlinearmethodsandtoolsfortheanalysisandpredictivemodelingofhigh- dimensionalphenomenaarenecessaryandmoreuseful. Fromanindustrialpointofview,forecastingtheelectricity/naturalgasconsump- tionofconsumersis yetanotherchallenge.Inordertobe abletoprovideaccurate forecasts on different horizons (from short term, the next hour, to middle term, several years) and at different levels of aggregation (from complete portfolio to localdemand),theseforecastshavetobedoneinanevolvingenvironment.Indeed, national demand is increasing, uses are changing, the number of available data is significantlygrowing(thankstosmartmeters),theindividualrenewableelectricity generationisbecomingimportant,andsmartgridsaredeveloping.Ontheotherside of electricity markets, renewables provide a bigger and bigger part of electricity generation.Forecastingwindorsolar powerisalso difficultbecauseoftheirinner variability,theneedforaccurateandlocalmeteorologicalforecastsatanyhorizon– from a few hours up to several days ahead – as well as the rapid evolution of operatinginstallations. OnJune5–7,2013,aninternationalworkshoponIndustryPracticesforForecast- ingwasheldinParis,France,organizedandsupportedbytheOSIRISDepartmentof EDFR&D,locatedinClamart,France.OSIRISstandsforOptimizationSimulation Risks and Statistics for energymarkets. The meeting was the second in the series v vi Preface of the WIPFOR conferencesand was attended by several researchers(academics, industrial and professionals, and other interested parties) from several countries. Followingtradition,boththeoreticalstatisticalresultsandpracticalcontributionsof this active field of statistical research and on forecasting issues in a fast evolving industrialenvironmentwerepresented.Theprogramandabstractsareavailableon theconferencewebsite(http://conferences-osiris.org/wipfor/13-Main-page). Theeditorsofthisvolumehopethattheselecturenotesreflectthebroadspectrum of the conference, as it includes 16 articles contributed by specialists in various areas in this field. The material compiled is fairly wide in scope and ranges from the development of results on forecasting in industry and in time series, on nonparametricandfunctionalmethods,ononlinemachinelearningforforecasting, andontoolsforhigh-dimensionalandcomplexdataanalysis. The articles are arranged and numbered in alphabetical order by author rather thansubjectmatter.Papers1,3,5,and9arededicatedtononparametrictechniques for short-term load forecasting in the industry and include classical curve linear regression, sparse functional regression based on dictionaries, as well as a new estimation procedure based on iterative bias reduction. Papers 11 and 14 focus on electrical system changes: the first is dedicated to large-scale electrical load simulationforsmartgridsusingGAMmodeling;thesecondfocusesonspace-time trajectories of wind power generation including parameterized precision matrices based on a Gaussian copula approach. Paper 7 provides flexible and dynamic modelingofdependenciesviacopulas.Papers6,8,and13exploredifferentaspects ofonlinelearningrangingfromthemostoperationalforonlineresidentialbaseline estimation to the more theoretical, which focuses on oracle boundsfor prediction errorsrelatedtoaggregationstrategies.Thethirdoneisdedicatedtotheaggregation of experts proposing some resampling ideas to enlarge the basic family of the so-calledexperts.Papers2,4,7,12,and16studysomegeneralapproachestohigh- dimensional and complex data (inference in high-dimensional models, graphical modelsandmodelselection,adaptivespotvolatilityestimationforhigh-frequency data, functionalclassification and prediction).Finally, papers10 and 15 deal with somespecialtopicsin time series, namely,modelingandpredictionof time series arisingonagraphandoptimalreconciliationofcontemporaneoushierarchicaltime seriesforecasts. We would like to address our gratitude to the keynote speakers and all the contributors for accepting to participate in these Lecture Notes, and we greatly appreciatethetimethattheyhavetakentopreparetheirpapers. To conclude, we would like to acknowledge the following distinguished list of reviewers who helped improve the papers by providing general and focused feedback to the authors: U. Amato, R. Becker, R. Cao, J. Cugliari, G. Fort, P. Fryzlewicz, F. Gamboa, B. Ghattas, I. Gijbels, S. Girard, E. Gobet, Y. Goude, R. Hyndman,I.Kojadinovic,S.Lambert-Lacroix,J.-M.Loubes,E.Matzner-Løber,G. Oppenheim,P.Pinson,G.Stoltz,A.Verhasselt,A.Wigington,andQ.Yao.Wethank themagainfortheirwork,dedication,andprofessionalexpertise. Preface vii Finally,tothe editorialofficeofSpringer-Verlagandparticularlyto EvaHiripi, wearegratefulfortheirefficientassistance. Grenoble,France AnestisAntoniadis Clamart,France XavierBrossat Paris,France Jean-MichelPoggi Paris,June2014. Contents ShortTermLoadForecastingintheIndustryforEstablishing ConsumptionBaselines:AFrenchCase ...................................... 1 José Blancarte, Mireille Batton-Hubert, Xavier Bay, Marie-AgnèsGirard,andAnneGrau ConfidenceIntervalsandTestsforHigh-DimensionalModels: ACompactReview............................................................... 21 PeterBühlmann ModellingandForecastingDailyElectricityLoadviaCurve LinearRegression................................................................ 35 HaeranCho,YannigGoude,XavierBrossat,andQiweiYao ConstructingGraphicalModelsviatheFocusedInformation Criterion .......................................................................... 55 GerdaClaeskens,EugenPircalabelu,andLourensWaldorp FullyNonparametricShortTermForecastingElectricity Consumption ..................................................................... 79 Pierre-AndréCornillon, Nick Hengartner,Vincent Lefieux, andEricMatzner-Løber ForecastingElectricityConsumptionbyAggregatingExperts; HowtoDesignaGoodSetofExperts.......................................... 95 PierreGaillardandYannigGoude FlexibleandDynamicModelingofDependenciesviaCopulas ............. 117 IrèneGijbels,KlausHerrmann,andDominikSznajder OnlineResidentialDemandReductionEstimationThrough ControlGroupSelection......................................................... 147 LeslieHatton,PhilippeCharpentier,andEricMatzner-Løber ix
The list of books you might like
