% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@PHDTHESIS{Hecht:974102,
author = {Hecht, Christopher},
othercontributors = {Sauer, Dirk Uwe and Marinelli, Mattia},
title = {{U}sage overview, prediction, and siting optimization for
electric vehicles public charging infrastructure with
machine learning and big data methods},
volume = {171},
school = {Rheinisch-Westfälische Technische Hochschule Aachen},
type = {Dissertation},
address = {Aachen},
publisher = {RWTH Aachen University},
reportid = {RWTH-2023-11220},
series = {Aachener Beiträge des ISEA},
pages = {1 Online-Ressource : Illustrationen},
year = {2023},
note = {Veröffentlicht auf dem Publikationsserver der RWTH Aachen
University 2024. - Überarbeitete Auflage mit Korrektur von
Abbildungen verfügbar; Dissertation,
Rheinisch-Westfälische Technische Hochschule Aachen, 2023},
abstract = {Anthropogenic climate change caused by greenhouse gas
emissions is among the biggest challenges that humanity must
tackle in the 21st century. We must reduce emissions rapidly
with the goal of reaching net-zero emissions by the middle
of the century as stated in the Paris Agreement. Looking at
the last years, mobility stands out as particularly hard to
decarbonize. For road traffic, electromobility is a path to
drive down emissions at the required scale and speed. To
realise this transition to electric drivetrains, charging
infrastructure is required. Next to private recharging
opportunities, public charging infrastructure is essential
to ensure that vehicles can also be recharged while away on
the road or while travelling longer distances. This
dissertation applies machine learning and big data methods
to improve the operation of charging infrastructure. Based
on extensive field datasets, we show typical usage patterns
for public charging infrastructure such as chargers being
occupied predominantly during the day and in urban centres.
Alternating current chargers are used more during the week
while direct current chargers are used more on the weekend.
These patterns are generated by users that can be grouped
into the categories “resident”, “commuter”, and
“opportunity user”. Commuters charge during the day and
more frequently in industrial areas as compared to the other
groups. Opportunity chargers also charge mostly during the
day but remain at the station for shorter periods and visit
stations much less frequently. Both groups drive the demand
spike during the day in the urban and industrial areas. In
contrast, residents generate demand in urban and suburban
environment and lean towards charging in the evening. The
combination of the groups leads to distinct patterns in the
various area types: Industrial areas experience a single
charging peak during the day and mostly on weekdays. Urban
areas show a strong peak during the day from commuters and
opportunity users working or visiting the areas. A second
peak is generated in the early evening from the returning
residents. Suburban areas show a similar pattern, but since
less people frequent these during the day, the daytime peak
is lower as in urban areas. Lastly, uninhabited areas are
mostly used by fast-chargers which are practically only used
by opportunity users coming in during the day.The machine
learning methods introduced in this dissertation are able to
predict usage patterns in time and space. For the prediction
in time, the tools introduced predict whether an existing
station will be occupied to at least $50\%$ with an accuracy
of $95\%$ and a Mathews Choice Coefficient of 0.8. The
algorithm further categorises occupation likelihood in the
terms “very low” to “very high”. This allows
navigation system operators, charge point operators, energy
suppliers, grid operators, and other actors to prepare the
infrastructure for possible demand spikes and to divert
users whose charging demand is flexible. Concerning the
prediction in space, the introduced regression model
achieves an R2 of 0.18 with practically all features having
a p-value of less than 0.05. The comparatively low R2
indicates that the model cannot explain the full complexity
of charging station usage but comes close given that many
unobservable factors play a role in charging station
occupation. Studies relying on large-scale data such as
provided in this dissertation have the benefit of providing
a representative and empirical picture of the world.
Model-based approaches are always subject to assumptions
that may not correspond with reality. To the best of my
knowledge, no other studies exist at the beginning of 2023
for Germany and few for other countries that provide the
given level of detail and accuracy.},
cin = {618310 / 614500},
ddc = {621.3},
cid = {$I:(DE-82)618310_20140620$ / $I:(DE-82)614500_20201203$},
typ = {PUB:(DE-HGF)11 / PUB:(DE-HGF)3},
doi = {10.18154/RWTH-2023-11220},
url = {https://publications.rwth-aachen.de/record/974102},
}
guest ::