Enhanced Machine Learning and Data Mining Methods for Analysing Large Hybrid Electric Vehicle Fleets based on Load Spectrum Data

192 Pages·2018·4.13 MB·English

by  Philipp Bergmeir (auth.)

Wissenschaftliche Reihe
Fahrzeugtechnik Universität Stuttgart
Philipp Bergmeir
Enhanced Machine Learning
and Data Mining Methods
for Analysing Large Hybrid
Electric Vehicle Fleets based
on Load Spectrum Data
Wissenschaftliche Reihe
Fahrzeugtechnik Universität Stuttgart
Reihe herausgegeben von
M. Bargende, Stuttgart, Deutschland
H.-C. Reuss, Stuttgart, Deutschland
J. Wiedemann, Stuttgart, Deutschland
Das Institut für Verbrennungsmotoren und Kraftfahrwesen (IVK) an der Universi-
tät Stuttgart erforscht, entwickelt, appliziert und erprobt, in enger Zusammenarbeit
mit der Industrie, Elemente bzw. Technologien aus dem Bereich moderner Fahr-
zeugkonzepte. Das Institut gliedert sich in die drei Bereiche Kraftfahrwesen, Fahr-
zeugantriebe und Kraftfahrzeug-Mechatronik. Aufgabe dieser Bereiche ist die Aus-
arbeitung des Themengebietes im Prüfstandsbetrieb, in Theorie und Simulation.
Schwerpunkte des Kraftfahrwesens sind hierbei die Aerodynamik, Akustik (NVH),
Fahrdynamik und Fahrermodellierung, Leichtbau, Sicherheit, Kraftübertragung
sowie Energie und Thermomanagement – auch in Verbindung mit hybriden und
batterieelektrischen Fahrzeugkonzepten.
Der Bereich Fahrzeugantriebe widmet sich den Themen Brennverfahrensent-
wicklung einschließlich Regelungs- und Steuerungskonzeptionen bei zugleich
minimierten Emissionen, komplexe Abgasnachbehandlung, Aufladesysteme und
-strategien, Hybridsysteme und Betriebsstrategien sowie mechanisch-akustischen
Fragestellungen.
Themen der Kraftfahrzeug-Mechatronik sind die Antriebsstrangregelung/Hybride,
Elektromobilität, Bordnetz und Energiemanagement, Funktions- und Softwareent-
wicklung sowie Test und Diagnose.
Die Erfüllung dieser Aufgaben wird prüfstandsseitig neben vielem anderen unter-
stützt durch 19 Motorenprüfstände, zwei Rollenprüfstände, einen 1:1-Fahrsimulator,
einen Antriebsstrangprüfstand, einen Thermowindkanal sowie einen 1:1-Aero-
akustikwindkanal.
Die wissenschaftliche Reihe „Fahrzeugtechnik Universität Stuttgart“ präsentiert
über die am Institut entstandenen Promotionen die hervorragenden Arbeitsergeb-
nisse der Forschungstätigkeiten am IVK.
Reihe herausgegeben von
Prof. Dr.-Ing. Michael Bargende Prof. Dr.-Ing. Jochen Wiedemann
Lehrstuhl Fahrzeugantriebe, Lehrstuhl Kraftfahrwesen,
Institut für Verbrennungsmotoren und Institut für Verbrennungsmotoren und
Kraftfahrwesen, Universität Stuttgart Kraftfahrwesen, Universität Stuttgart
Stuttgart, Deutschland Stuttgart, Deutschland
Prof. Dr.-Ing. Hans-Christian Reuss
Lehrstuhl Kraftfahrzeugmechatronik,
Institut für Verbrennungsmotoren und
Kraftfahrwesen, Universität Stuttgart
Stuttgart, Deutschland
Weitere Bände in der Reihe http://www.springer.com/series/13535
Philipp Bergmeir
Enhanced Machine
Learning and Data
Mining Methods
for Analysing Large
Hybrid Electric Vehicle
Fleets based on Load
Spectrum Data
Philipp Bergmeir
Stuttgart, Germany
Zugl.: Dissertation Universität Stuttgart, 2017
D93
Wissenschaftliche Reihe Fahrzeugtechnik Universität Stuttgart
ISBN 978-3-658-20366-5 ISBN 978-3-658-20367-2 (eBook)
https://doi.org/10.1007/978-3-658-20367-2
Library of Congress Control Number: 2017961103
Springer Vieweg
© Springer Fachmedien Wiesbaden GmbH 2018
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms 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 specific 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. The publisher remains neutral with
regard to jurisdictional claims in published maps and institutional affiliations.
Printed on acid-free paper
This Springer Vieweg imprint is published by Springer Nature
The registered company is Springer Fachmedien Wiesbaden GmbH
The registered company address is: Abraham-Lincoln-Str. 46, 65189 Wiesbaden, Germany
„ForTina.“
Acknowledgements
I would like to thank all those people who made this thesis possible and an
unforgettableexperienceforme.
First,IwouldliketoacknowledgemygratitudetoProf.Dr.-Ing.MichaelBar-
gende,myadvisorattheUniversityofStuttgart,forhisgreatsupportandhis
permanentcommitmentinthedoctoralprogram“PromotionskollegHYBRID”
in which I was participating. Without him, I probably would not have been
given the chance to write this thesis as part of a collaboration between the
UniversityofStuttgart,theEsslingenUniversityofAppliedSciences,andthe
DaimlerAG.
IalsooweProf.Dipl.-Ing.JürgenNonnast,myadvisorattheEsslingenUniver-
sityofAppliedSciences, agreatdebtofgratitude: Hewasalwayswillingto
listentomeandhelpedmewithtechnicalaswellasorganizationalquestions.
Healsogavemetheopportunitytodesignandtogivemyownlecturesabout
DataMiningattheDepartmentofInformationTechnology.
Next, I would like to thank Prof. Dr.-Ing. Oliver Sawodny for his interest in
thisworkandforjoiningthedoctoralcommitee.
IwouldliketoexpressmydeepsenseofgratitudetoDr.ChristofNitsche,my
advisoratDaimlerAG,forhiscontinuousadviceandencouragementhegave
tomethroughoutthefouryearsIspentonthisthesis. Ithankhimforhispa-
tience, motivation, and enthusiasm. His systematic guidance and experience
helpedmeinallthetimeofresearchandwritingofthisthesis.
MyverysincerethanksalsogoestothewholedepartmentRD/PGHatDaimler
AG for offering me the opportunity to work on a thrilling industrial research
projectregardinghybridelectricvehicles. Namely,IwanttothankPeterAnt-
ony,myteamleaderatDaimlerAG,andtheheadsofthedepartmentRD/PGH
duringmyresearchtime, Dr.UweKellerandJochenStrenkert, fortheirsup-
port,especiallyinorganizationalmatters.
For many fruitful discussions and for supporting me in programming tasks,
myspecialthanksgoestomycolleaguesGauthamRaju,VishalRatra,andKir-
ankumarReddyfromMercedes-BenzResearchandDevelopmentIndia.
VIII Acknowledgements
Furthermore, IalsowanttoexpressmygratitudetotheMinistryforScience,
ResearchandArtsBaden-Württembergforfundingthedoctoralprogram“Pro-
motionskollegHYBRID”.
I am thankful to Dr. Andreas Theissler for being my co-lecturer in the two
courses “Introduction to Data Mining” and “Intelligent Data Analytics”. For
sure,Iamgoingtomissourexcellentcollaboration.
For their great technical support regarding the usage of the computer cluster
of the Esslingen University of Applied Sciences, I want to thank Dr. Adrian
ReberandAlexandruSaramet.
Moreover, I thank my friends and former, fellow Ph.D. students Dr. Daniel
GörkeandDr.AndreasHaagforallthestimulatingtechnicalandnon-technical
discussionswehadaswellasfortheathleticismweperformedinourleisure
times.
I am indebted to my cousin, Dr. Christoph Bergmeir, for his helpful advices,
inparticularregardingtheprocessofpublishingscientificpapers.
None of this would have been possible without the love and patience of my
family. That is why I take this opportunity to express the profound gratitude
from my deep heart to my parents, my grandparents, my sister, my brother-
in-law, mynieceLuisa, mynephewLukas, mysister-in-law, andmyparents-
in-law for their unconditional support throughout my whole life. They never
doubted any of my decisions, they helped me to clear my mind in stressful
times,andIamverygladtoknowthattheywillalwaysstandbyme,ingood
aswellasinbadtimes.
Finally, but most importantly, I would like to thank my beloved wife Tina.
Withoutherendlesssupport, itwouldhavebeenimpossibleformetomaster
thechallengeofwritingthisthesis. DearTina,thankyouforunderstandingall
ofmy decisionsandfor sticking byme, inspite ofthegeographical distance
betweenusduringthelastfewyears. ThankyouforlovingmejustthewayI
am.
Augsburg PhilippBergmeir
Contents
Acknowledgement.......................................... VII
ListofFigures .............................................. XIII
ListofTables ............................................... XVII
Acronyms................................................... XXI
Symbols..................................................... XXIII
Abstract ..................................................... XXIX
Kurzfassung................................................. XXXI
1 Introduction............................................. 1
1.1 Aims .............................................................. 2
1.2 Relatedwork...................................................... 2
1.3 Ownpublications................................................. 4
1.4 Outline............................................................ 5
2 Datafoundation ........................................ 7
2.1 Datasources...................................................... 7
2.1.1 On-boarddata: loadspectrumdata....................... 8
2.1.2 Off-boarddata: workshopdata ........................... 13
2.2 Preprocessingofdata ............................................ 15
2.3 Real-worlddatasets .............................................. 16
2.4 Conclusion........................................................ 17
3 Classifyingcomponentfailuresofavehiclefleet.... 19
3.1 Fundamentalsofclassification................................... 19
3.1.1 PerformanceMeasures.................................... 22
X Contents
3.1.2 The“classimbalanceproblem”........................... 23
3.2 Classificationmethods........................................... 25
3.2.1 Supportvectormachine(SVM) .......................... 25
3.2.2 Classificationtree ......................................... 33
3.2.3 Randomforest(RF)....................................... 38
3.2.4 Obliquerandomforest(ORF) ............................ 41
3.3 Fundamentalsoffeatureselection............................... 47
3.3.1 Variableimportanceintree-basedclassifiers ............ 50
3.3.2 Recursivefeatureelimination(RFE)..................... 53
3.4 AnewRFbasedclassificationandfeatureselectionframework 55
3.5 Casestudy: Classifyingcomponentfailuresofahybridcar
battery............................................................. 57
3.5.1 Parameteroptimization.................................... 58
3.5.2 Results ..................................................... 65
3.6 Conclusion........................................................ 81
4 Visualizingdifferentkindsofvehiclestressand
usage ..................................................... 83
4.1 Distanceanddissimilaritymeasures ............................ 84
4.1.1 Euclideandistance......................................... 84
4.1.2 Randomforestdissimilarity .............................. 85
4.2 Dimensionalityreductionmethods.............................. 87
4.2.1 PrincipalComponentsAnalysis .......................... 89
4.2.2 SammonMapping......................................... 90
4.2.3 LocallyLinearEmbedding................................ 91
4.2.4 Isomap ..................................................... 93
4.2.5 t-DistributedStochasticNeighbourEmbedding ......... 93
4.3 Casestudy: Dependenceofvehicleusageonoperatingcountry 97
4.3.1 Preprocessingandparametrization....................... 98
4.3.2 Results ..................................................... 99