p25 / Preprint
Federated Learning for Early Prediction of EV Charging Demand
arXiv:2605.04993 · 2026
In plain language
The total energy of an EV charging session is estimated from plug-in information and its first minutes, while intervention is still possible.
Research context
Early estimates can support charging planning without waiting for a completed session.
Approach
ACN data from Caltech combines session metadata and early charging measurements. Station-level partitions capture heterogeneity in federated model comparisons.
Findings
Federated predictions approach centralized performance while keeping data in the depot. Useful demand estimates are available early with limited session information.
Summary based on abstract.