Machine learning approaches for the prediction of public EV charge point flexibility

The increasing number of electric vehicles on the road can play a key role as a source of flexibility for a reliable power system operation. Charge point operators, in particular, can adjust individual electric vehicle charging loads to provide system operators with aggregated energy flexibility, e.g. for congestion management, ancillary services or greenhouse gas emission reduction. However, managing individual charging sessions requires information about the expected session duration and energy demand, which are not available at the beginning of a session. In this work, a novel predictive workflow based on two causality-informed machine learning approaches with different levels of generalization is proposed to predict individual session duration and energy demand. Our key contributions include the development of a cluster-based predictive model for charge points and a user-based predictive model to capture individual charging behaviours, and the comparison of these models using a large-scale, real-world dataset. The proposed approaches were tested on real charging data provided by TotalEnergies, showing that considering user-specific charging behaviours enhances the accuracy performance by 16.1% and 37.9% for predictions of session duration and energy demand, respectively. By leveraging clustering and feature selection techniques, accounting for charge point- and user-specific charging patterns, and utilizing a large-scale real-world charging dataset, the proposed predictive workflow enables a comprehensive comparison of machine learning techniques in terms of accuracy performance when predicting public electric vehicle charge point flexibility.