top of page

Blog 11 -  Predicting EV charging duration using machine learning 

Musa, 18 July 2024

The first publication for my PhD research is a conference paper, with title 'Predicting EV charging duration using machine
learning and charging transactions at three sites' . 

Predicting the duration of electric vehicle (EV) charging is relevant for the swift integration of EVs into the grid. It informs EV charging park operators on when to expect peak demand, and the time it takes to charge EVs. This paper describes an approach, based on three machine learning models, to predict the EV charging duration at three public charging sites
in the city of Leeds (United Kingdom).

 

The prediction is based on the use of a dataset comprising a total of 7271 charging sessions for the year 2019. The figure below shows the methodology adopted in the paper, highlighting the main steps undertook in the process: data preprocessing, feature engineering, separation of the data set in two groups (for training and testing), prediction model, evaluation and results. 

 

graph.JPG
Maching learning prediction loadflow presented in the paper. 

In the first part of the paper, the characteristics of the considered dataset are described. The second part of the paper shows that accurate prediction performance can be achieved using features included in the charging transactions exclusively. 

My paper showed that the predictive performance at the Temple Green site which was one of the three charging sites explored is the best because its data has fewer outliers and has many identical users frequenting the site, and fewer random users than the other two charging sites. This is one of the advantages of using machine learning techniques.

Moreover, my paper underscored the importance of smart charging solutions, leveraging data analytics and predictive algorithms to optimize charging schedules so that charging operators are well-informed ahead of time regarding peak demand and charging duration of EV users. By knowing the expected charging duration, peak demand would be minimized, the strain on the grid could be reduced, and a seamless charging experience for EV owners ensured.

A copy of the accepted paper can be accessed at King's repository and on IEEE Explorer.  

My plan is now to continue dwelving into these topics with more details, and in particular I am considering new datasets, the use of clustering techniques, and addressing the impact of charging on the electrical grid. 

Contact: zulkiflu.musa_sarkin_adar@kcl.ac.uk


 

bottom of page