Study on best algorithm to predict idle time in EV charging

It is becoming a widely discussed aspect of charging electric cars: Idle time, the time an EV is connected to a charging station without charging. The factor idle time is directly related to the sizing of the charging infrastructure, hence its cost or availability and the possibilities for Smart Charging. A new study on idle time by Alexandre Lucas, Ricardo Barranco (both EU Joint Research Centre) and Nazir Refa (ElaadNL) is now published in scientific journal Energies: EV Idle Time Estimation on Charging Infrastructure, Comparing Supervised Machine Learning Regressions.

The time an EV takes to charge depends on its initial/final state of charge (SoC) and the power being supplied to it. The problem however is to estimate the time the EV remains parked after charging (idle time), as it depends on many factors which simple statistical analysis cannot tackle. In this study the researchers applied supervised machine learning to an EV charging dataset from the public chargers located in the Netherlands. Three regression algorithms (Random Forest, Gradient Boosting, and XGBoost) have been analysed in order to identify the most accurate and main influencing parameters constructing a model for idle time prediction.

Highest accuracy

The best performing model is XGBoost proving itself by having the highest accuracy. The parameters influencing the model the most are: the time of day in which the charging sessions start, and the total energy supplied. Partial dependencies of variables and model performances are also presented in the study and implications on public policies are discussed.

Looking at the intraday hour’s parameter Lucas, Barranco and Refa observed that sessions starting from 5 PM onwards (mainly ‘night’ chargers) have the highest impact on the values of predicted idle time. While the charging events starting around 2 PM (‘visitors’ with mainly shorter sessions) seems to have less impact in the idle time prediction.  

Joint research activities

The model can provide useful information for EV users, policy makers and network owners to better manage the network, targeting specific variables. For example, using Smart Charging possibilities.

ElaadNL and JRC working in close collaboration on EV (data) research from 2015 onwards. This study can be seen as a product of this collaboration. Both organizations aim to intensify their joint research activities in the near future on wide aspects related to EVs and electricity grids.    

Download the report at MDPI

Our latest news