Energy Time Series Forecasting: Efficient and Accurate Forecasting of Evolving Time Series from the Energy Domain

241 Pages·2015·6.79 MB·English

by  Lars Dannecker (auth.)

Energy Time Series Forecasting
Lars Dannecker
Energy Time Series
Forecasting
Effi cient and Accurate Forecasting
of Evolving Time Series from
the Energy Domain
Lars Dannecker
Dresden, Germany
Doctorate at the Technische Universität Dresden, 20.11.2014
Original title: Efficient and Accurate Forecasting of Evolving Time Series from the
Energy Domain
ISBN 978-3-658-11038-3 ISBN 978-3-658-11039-0 (eBook)
DOI 10.1007/978-3-658-11039-0
Library of Congress Control Number: 2015947268
Springer Vieweg
© Springer Fachmedien Wiesbaden 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or
part of the material is concerned, speci(cid:191) cally the rights of translation, reprinting, reuse of illus-
trations, recitation, broadcasting, reproduction on micro(cid:191) lms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by
similar or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a speci(cid:191) c statement, that such names are
exempt from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained
herein or for any errors or omissions that may have been made.
Printed on acid-free paper
Springer Vieweg is a brand of Springer Fachmedien Wiesbaden
Springer Fachmedien Wiesbaden is part of Springer Science+Business Media
(www.springer.com)
TomywifeUlrikeandmydaughterLuisa.
Preface
Continuous balancing of electric power consumption and production is a funda-
mental prerequisite for the stability and efficiency of electricity grids. This bal-
ancing task requires accurate forecasts of future electricity demand and supply
at any point in time. For this purpose, today’s energy data management systems
(EDMS)typicallyusequantitativemodels—calledforecastmodels—thatalready
provideaccuratepredictions.However,recentdevelopmentsintheenergydomain
such as real-time intra-day trading and the integration of more renewable energy
sourcesalsorequiremoreefficientforecastingcalculationsandarapidprovision-
ingofforecastingresults.Furthermore,today’sEDMSsfulfillanumberofdiffer-
ent tasks, each exhibiting different requirements for the calculation of forecasts
with respect to runtime and accuracy. Thus, it is necessary to flexibly adapt the
forecasting process with respect to the needs of the current requests. In contrast,
currentlyemployedforecastingapproachesarerathertime-consumingandinflexi-
ble.Onereasonistheveryexpensiveestimationoftheforecastmodelparameters,
involvingalargenumberofsimulationsinasearchspacethatincreasesexponen-
tialwiththenumberofparameters.
We tackle these new requirements by introducing a novel online forecasting
processthataimstoimprovetheforecastingcalculationefficiencyandtoprovide
forecasting process adaptability. For this purpose, the online forecasting process
employs forecast model materialization in conjunction with flexible and fast pa-
rameter estimation to rapidly provide accurate forecasts that are iteratively im-
proved over time. EDMSs may subscribe to the online forecasting process to re-
trieve improvements found during the process execution. In addition, they can
adapt the progression of the forecast calculation by defining runtime constraints
andaccuracytargets.Withthat,weareabletoequallyserverrequeststhatrequire
resultsinalimitedamountoftimeorthattargetthebestpossibleaccuracy.
The online forecasting process is complemented by further optimizations on
thelogicalaswellasonthephysicallayer.Ouroptimizationsonthelogicallayer
viii Preface
improvetheefficiencyoftheparameterestimationindependentlyofthedataorga-
nization and the employed forecast model. As a first approach, we introduce our
context-awareforecastmodelrepositorythatmaterializespreviouslyusedforecast
modelsandtheirparametersinconjunctionwithinformationaboutthetimeseries
contextthatwasvalidduringthetimethemodelwasused.Wemaythenprovide
appropriate starting points for future forecasting calculations by reusing models
thatproducedaccurateresultsinacontextsimilartothecurrentone.Furthermore,
forsomeusecases,itisbeneficialtoconsidercontextinformationdirectlywithin
the forecast models. Especially when predicting renewable supply, information
abouttheweatherareveryimportant.However,includingcontextinformationtyp-
icallymeanstoaddfurtherparameterstotheforecastmodel,whichincreasesthe
effortsfortheparameterestimation.Tosolvethisissue,weintroduceanintegra-
tionframeworkthatoptimizesthehandlingofcontextinformationandreducesthe
additional efforts when considering them. Finally, we improve the calculation of
forecastsinhierarchicalenvironments.Insteadofsimplyaggregatingtimeseries,
we propose a forecast model aggregation that eliminates the need for estimating
theforecastmodelparametersonhigherhierarchicallevels.
Ourphysicaloptimizationsaimtodirectlyprovideanefficientwayforforecast
models to access time series values. For this purpose, we introduce an access-
pattern-aware storage approach that exploits the memory access patterns of the
usedforecastmodelstophysicallylayoutthedataforsequentialaccessandhigh
spatiallocality.Withthat,wesubstantiallyreducethenegativeinfluenceofmem-
orylatencyandbandwidth,whileatthesametimeimprovingtheutilizationofthe
different cache levels. In addition, we propose a special parallelization approach
formulti-equationforecastmodels.
Overall, with the help of our online forecasting process in conjunction with
the optimizations on the logical and on the physical layer, we target to enable
accurate forecasting of evolving time series considering the new requirements of
thechangingelectricitymarket.
Acknowledgements
Firstandforemost,IwouldliketoexpressmydeepestgratitudetomyadvisorProf.
Dr. Wolfgang Lehner for the opportunity to pursue my dissertation project and
forhiscontinuoussupportalongtheway.Ithankhimfortheveryinterestingand
challengingtopicandforalltheadvicehegaveme.Wolfgangwasalwaysavailable
foradiscussionandforprovidingfeedbacktomyideas.I’mverythankfulthatI
had the chance to leverage his deep knowledge in all areas of data management.
AlthoughIwasanexternalmemberofhisresearchgroup,Wolfgangneverletme
feelexternal,butrathertreatedmeasanintegralpart.Forme,Wolfgangisthiskind
of advisor who all PhD students should have. I really appreciate all the time and
guidanceyouprovidedduringthelastyears.Thankyousomuchforeverything.
I would also like to thank Dr. Matthias Bhm for acting as my co-mentor in
thefirsttwoyears.Hewasdeeplyinvolvedinshapingthetopicofthethesisand
creatingthefirstscientificresults.Matthiasdevotedalotoftimefordiscussions,
forprovidingadviceandguidance,andforgivingconstructivefeedbacktomany
papers.Ilearnedsomuchfromhim,especiallywithrespecttoconductingresearch,
creatingpublicationsandevaluatingmyideas.Matthias,thisthesiswouldnothave
beenpossiblewithoutyou.Ireallyappreciateyourdedicationandendlesssupport.
Writing this thesis was greatly supported by the SAP SE and especially the
SAPlocationinDresden.SAPemployedmeasaPhDstudentandthus,gaveme
the chance to start my dissertation project. I want to especially thank Dr. Gregor
Hackenbroichforservingasmymentoraswellasforhisconfidence,hispatience,
andhisbeliefinmethroughouttheentiretime.Inmanydiscussions,Gregorpro-
videdmewithveryhelpfulinputanddirectionsforvariousaspectsofmythesis.
Thankyouverymuch,I’mreallygratefulforyoursupport.Further,Iwouldliketo
thankmycolleagueDr.PhilippRschforbeingaco-authorofmanypapersandthe
endlessdiscussionsaboutmysometimesstrangeideas.Withhisdeepknowledge,
he provided great guidance and helped forming several approaches and topics of
x Acknowledgements
thisthesis.ThankyouPhilippfortakingsomuchtimeandendurance,especially
inrelationtothesecondhalfofmythesis.
Furthermore,IwouldliketothankProf.ChristianJensenforco-referringthis
thesis and for many helpful comments and advice. I really appreciate his great
hospitalityduringmyvisitsinAarhusandAalborg.I’malsogratefultoProf.Dr.
DominikMstforservingasmyFachreferent.
Special thanks goes to my students Robert Schulze, Elena Vasilyeva, Robert
Lorenz,andGordonGaumnitzforcontributingtomyresearchandforhelpingto
implementthepEDMprototype.ItwasapleasureworkingwithyouandI’mvery
proudthatallofyouachievedsuchgoodresultsinyourfinaltheses.
IamverythankfultoallmycolleaguesatSAPSEandtheTechnischeUniver-
sitt Dresden who helped me with their input and discussions. I would especially
like to thank: Dr. Gregor Hackenbroich and Dr. Philipp Rsch for proof-reading
my thesis and for providing very constructive comments; Dr. Ulrike Fischer for
working with me in the MIRABEL project and the fruitful discussions we had;
Konrad,Katrin,Andreas,Henrike,andAlexandrforbeinggreatroommates;Ines
and Annette for helping me with (and enduring) my numerous requests in their
role as team assistants; my fellow researchers in the database technology group
Dirk, Maik, Martin, Hannes, Tim, Katrin, Julian, Thomas, Tobias, Tomas, Till,
Claudio,Kai,Frank,Robert,andElenaforyouracceptanceandafruitfulresearch
environment;theremainingcolleaguesatSAPSEespeciallyKay,Martin,Philipp,
Michael,Marcus,David,Robert,Dan,Ivan,andKarimforbeinggoodcolleagues
andforprovidingagreatworkingatmosphere.
Finally, this thesis would not have been possible without the constant support
and motivation from my family and friends. First and foremost, I would like to
thankmybelovedwifeUlrikeforalwaysbeingthereforme,forencouragingme,
andforacceptingthenumeroustimesIhadtoworkathome.Yourloveandcom-
mitmentiswhathasalwaysmotivatedme.Ialsowanttothank:MydaughterLuisa
forcheeringmeupandforremindingmeaboutthemostimportantthingsinlife;
myparentsJuttaandFalkaswellasmybrotherFrankforcontinuouslysupporting
me, for permanently believing in me, and for helping me out of the tough situ-
ations where I had a lot of doubts; Kay, Martin, Jan, Gregor, Stefan, and all my
otherfriendsforyourfriendshipandthepatienceyouhadwithme.Thankyou!I
willdedicatemoretimetoallofyouinthefuture.
LarsDannecker
Dresden,12.August2014
Contents
1 Introduction.................................................. 1
2 TheEuropeanElectricityMarket:AMarketStudy ............... 11
2.1 CurrentDevelopmentsintheEuropeanElectricityMarket ....... 12
2.1.1 StructureoftheEuropeanElectricityMarket ............ 12
2.1.2 DevelopmentofRenewableEnergySourcesinEurope
andGermany....................................... 13
2.1.3 ImpactofVolatileRenewableEnergySources ........... 17
2.1.4 HowtoKeeptheElectricityGridinBalance ............ 20
2.1.5 ExtendingtheTransmissionGridandEnergyStorage..... 25
2.1.6 Demand-SideManagementandDemand-Response....... 30
2.1.7 ChangesontheEuropeanElectricityMarket ............ 32
2.1.8 Improvements in Forecasting Energy Demand and
RenewableSupply .................................. 37
2.2 TheMIRABELProject:ExploitingDemandandSupplySide
Flexibility................................................ 41
2.2.1 Flex-Offers ........................................ 41
2.2.2 ArchitectureofMIRABEL’sEDMS ................... 43
2.2.3 BasicandAdvancedUse-Case ........................ 45
2.3 Conclusion............................................... 46
3 TheCurrentStateofEnergyDataManagementandForecasting ... 49
3.1 DataCharacteristicsintheEnergyDomain .................... 50
3.1.1 SeasonalPatterns ................................... 51
3.1.2 Aggregation-Level-DependentPredictability ............ 53
3.1.3 TimeSeriesContextandContextDrifts ................ 56
3.1.4 TypicalDataCharacteristicsofEnergyTimeSeries....... 58
3.2 ForecastingintheEnergyDomain ........................... 59
3.2.1 ForecastModelswithAutoregressiveStructures ......... 59