Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Multivariate time series classification in time-sensitive environments using deep learning

Östergren, Hannes LU (2024) In Master's Theses in Mathematical Sciences FMAM05 20232
Mathematics (Faculty of Engineering)
Abstract
In this thesis, sensor data from substations and distribution centrals in Sweden and Germany is
analysed and used to predict the voltage environment which the device is within. The theory needed
for understanding the measured values is explained, and the notion of multivariate time series is
introduced along with the reason for using deep learning to solve the problem at hand. Several ways
of extracting features to visualize data and improve classification accuracy are introduced, and the
resulting plots are analyzed to give insight into what results can be expected from the training phase.
Furthermore, common model architectures and state-of-the-art models are explained, as well as why
they are suitable as comparators in the... (More)
In this thesis, sensor data from substations and distribution centrals in Sweden and Germany is
analysed and used to predict the voltage environment which the device is within. The theory needed
for understanding the measured values is explained, and the notion of multivariate time series is
introduced along with the reason for using deep learning to solve the problem at hand. Several ways
of extracting features to visualize data and improve classification accuracy are introduced, and the
resulting plots are analyzed to give insight into what results can be expected from the training phase.
Furthermore, common model architectures and state-of-the-art models are explained, as well as why
they are suitable as comparators in the thesis. An extensive comparison between algorithms and
deep learning models is then carried out to find the suitability of different models for the classification
task by comparing accuracy, inference time (forward propagation time) and storage space (model
size). Finally, the results and their implications are discussed along with the assumptions made while
collecting the data to give the reader an understanding of the results. Improvements on the method
used, issues that were encountered during the thesis and ideas for future work are also discussed. (Less)
Popular Abstract (Swedish)
Elektriker och servicepersonal vistas frekvent inom områden där högspänningskomponenter förekommer. Detta arbete undersöker möjligheten att utnyttja maskininlärning, mer specifikt djupinlärning som grund i ett system som varnar elektriker och servicepersonal om strömförande komponenter och ledningar.
Please use this url to cite or link to this publication:
author
Östergren, Hannes LU
supervisor
organization
course
FMAM05 20232
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Deep learning, Machine learning, Multivariate time series, time series, E-field, B-field, electric, magnetic, safety
publication/series
Master's Theses in Mathematical Sciences
report number
LUTFMA-3527-2024
ISSN
1404-6342
other publication id
2024:E10
language
English
id
9149054
date added to LUP
2024-03-12 11:36:49
date last changed
2024-03-12 11:36:49
@misc{9149054,
  abstract     = {{In this thesis, sensor data from substations and distribution centrals in Sweden and Germany is
analysed and used to predict the voltage environment which the device is within. The theory needed
for understanding the measured values is explained, and the notion of multivariate time series is
introduced along with the reason for using deep learning to solve the problem at hand. Several ways
of extracting features to visualize data and improve classification accuracy are introduced, and the
resulting plots are analyzed to give insight into what results can be expected from the training phase.
Furthermore, common model architectures and state-of-the-art models are explained, as well as why
they are suitable as comparators in the thesis. An extensive comparison between algorithms and
deep learning models is then carried out to find the suitability of different models for the classification
task by comparing accuracy, inference time (forward propagation time) and storage space (model
size). Finally, the results and their implications are discussed along with the assumptions made while
collecting the data to give the reader an understanding of the results. Improvements on the method
used, issues that were encountered during the thesis and ideas for future work are also discussed.}},
  author       = {{Östergren, Hannes}},
  issn         = {{1404-6342}},
  language     = {{eng}},
  note         = {{Student Paper}},
  series       = {{Master's Theses in Mathematical Sciences}},
  title        = {{Multivariate time series classification in time-sensitive environments using deep learning}},
  year         = {{2024}},
}