Thesis

Predicting electric vehicle charging station occupancy : model evaluation and usability insights

Downloadable Content

Download PDF
Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2025
Thesis identifier
  • T17455
Person Identifier (Local)
  • 201975263
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • Electric vehicles are increasingly adopted as an environmental complement to internal combustion vehicles through incentives provided by governments worldwide. However, despite rapid growth in certain market segments, concerns over range and the availability of charging infrastructure thwart further acceptance among the general public. A key challenge is the uncertainty around the unavailability of public charging stations, which in turn increases “range anxiety” for EV drivers. This thesis addresses the challenge of deep learning models’ performance in EV charging stations’ availability forecasting. A preliminary analysis of the charging behaviours by historical data samples and the quantitative survey with EV drivers was performed in order to extract the essential temporal and environmental factors affecting the station occupancy. This insight was used in investigating performances of several deep learning architectures for classification and regression tasks, particularly Long-Short Term Memory (LSTM), Gated Recurrent Unit (GRU), Temporal Convolutional Network (TCN), 1Dimensional Convolutional Network (1D-CNN), and an Ensemble model (comprising 1D-CNN and TCN). The experimental results revealed that classification models can offer superior occupancy status prediction performance compared to regression models, and the optimum performance is obtained using 1D-CNN and TCN. A new hybrid architecture combining TCN and bidirectional GRU (BiGRU) achieved enhanced classification accuracy and generalized the performance of the baseline models. A complementary quantitative and then qualitative investigation were also conducted to establish the preferences of electric vehicle owners concerning charging forecasting, and their level of trust in model forecasting. This research enhances the management of EV charging infrastructure by delivering precise and user-centric occupancy predictions. By mitigating range anxiety and increasing confidence in predictive systems, it facilitates informed decision-making, enhances the charging experience, and encourages broader worldwide use of electric vehicles.
Advisor / supervisor
  • Dunlop, Mark
Resource Type
DOI

Relations

Items

Thumbnail Title Date Uploaded Visibility Actions
file details PDF of thesis T17455 2025-09-09 Public Download