Predicting electric vehicle charging connector type availability : The effects of weather, traffic and holiday data
Niiranen, Minttu (2025)
2025
Teknis-luonnontieteellinen DI-ohjelma - Master's Programme in Science and Engineering
Tekniikan ja luonnontieteiden tiedekunta - Faculty of Engineering and Natural Sciences
Hyväksymispäivämäärä
2025-11-20
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-2025112010785
https://urn.fi/URN:NBN:fi:tuni-2025112010785
Tiivistelmä
Electric vehicles (EVs) are essential for reducing harmful emissions and greenhouse gases caused by traffic, as electric motors do not produce tailpipe emissions, unlike internal combustion engines. Reducing greenhouse gas emissions is crucial for mitigating climate change, protecting ecosystems, and creating a sustainable world for future generations.
The global market for EVs has continued to increase in recent years, but widespread EV adoption still faces significant challenges. One critical challenge is the state of public charging infrastructure, which faces inadequate coverage in certain regions, long charging times, and difficulty locating available chargers before battery depletion. More importantly, the availability of a suitable connector type, a type of socket or cable, is necessary to successfully charge an EV.
This thesis develops EV charger connector type level availability predictions with a one-week prediction horizon to address the challenge of locating compatible available chargers in advance and enable improved trip planning. The goal is to examine the effects of weather, traffic, and holiday data on availability predictions at charging locations in Finland, Australia, and the Netherlands.
The connector type level availability predictions utilize linear regression (baseline), XGBoost regression (ensemble method), and beta regression models (primary model), which are compared against the median predictions in the HERE EV product. Since availability data includes proportional values between 0 and 1, beta regression serves as the primary model. Beta regression is particularly advantageous for proportional data as it accommodates diverse distributional shapes, including skewed data, while constraining predictions to the valid range of proportional data.
The effects of additional features on availability data are examined by linear correlation analysis before incorporating the features into prediction models. The predictions were calculated for the first week of January 2025, which included holidays in each target country. Predictions included 1669-3299 location and connector type groups, and the performance was evaluated with RMSE, MAE and R-squared. The results revealed that holidays were overall the most effective feature for improving predictions, followed by traffic levels, while temperatures deteriorated performance. The effects on the mean RMSE and MAE values were modest, with improvements typically seen in the third or fourth decimal places. However, holidays improved 58-71% of the predictions, with up to 2.6-57% RMSE improvements in Finland. Traffic levels achieved more moderate 1-10% RMSE improvements. Temperatures deteriorated more than half of the predictions, but provided up to 2.6-54% RMSE improvements, demonstrating notable advantage in specific cases.
The results demonstrated the importance of holidays in enhancing predictions, as well as the challenge of applying regression models and additional features to all location and connector type groups. The results showed significant improvements in the subsets of predictions, although the overall effects were moderate. This thesis opens opportunities for future research, of which the most important is examining the possible common factors behind the deteriorated predictions.
The global market for EVs has continued to increase in recent years, but widespread EV adoption still faces significant challenges. One critical challenge is the state of public charging infrastructure, which faces inadequate coverage in certain regions, long charging times, and difficulty locating available chargers before battery depletion. More importantly, the availability of a suitable connector type, a type of socket or cable, is necessary to successfully charge an EV.
This thesis develops EV charger connector type level availability predictions with a one-week prediction horizon to address the challenge of locating compatible available chargers in advance and enable improved trip planning. The goal is to examine the effects of weather, traffic, and holiday data on availability predictions at charging locations in Finland, Australia, and the Netherlands.
The connector type level availability predictions utilize linear regression (baseline), XGBoost regression (ensemble method), and beta regression models (primary model), which are compared against the median predictions in the HERE EV product. Since availability data includes proportional values between 0 and 1, beta regression serves as the primary model. Beta regression is particularly advantageous for proportional data as it accommodates diverse distributional shapes, including skewed data, while constraining predictions to the valid range of proportional data.
The effects of additional features on availability data are examined by linear correlation analysis before incorporating the features into prediction models. The predictions were calculated for the first week of January 2025, which included holidays in each target country. Predictions included 1669-3299 location and connector type groups, and the performance was evaluated with RMSE, MAE and R-squared. The results revealed that holidays were overall the most effective feature for improving predictions, followed by traffic levels, while temperatures deteriorated performance. The effects on the mean RMSE and MAE values were modest, with improvements typically seen in the third or fourth decimal places. However, holidays improved 58-71% of the predictions, with up to 2.6-57% RMSE improvements in Finland. Traffic levels achieved more moderate 1-10% RMSE improvements. Temperatures deteriorated more than half of the predictions, but provided up to 2.6-54% RMSE improvements, demonstrating notable advantage in specific cases.
The results demonstrated the importance of holidays in enhancing predictions, as well as the challenge of applying regression models and additional features to all location and connector type groups. The results showed significant improvements in the subsets of predictions, although the overall effects were moderate. This thesis opens opportunities for future research, of which the most important is examining the possible common factors behind the deteriorated predictions.