2026

Aushev, Alexander; Anttila, Joel; Todorov, Yancho; Hentunen, Ari; Pihlatie, Mikko

The rapid expansion of electric vehicles (EVs) in response to stricter emissions targets presents formidable challenges for power systems, particularly in scaling EV charging infrastructure to meet growing demands from heavy-duty fleets. Such demands are shaped by complex spatio-temporal interdependencies, such as weather conditions, traffic density, routes, and charging infrastructure, leading to imprecise charging demand predictions by the existing models that do not fully address all factors. This study introduces the Weather Traffic Routes and Chargers (WeTRaC), a predictive framework that unifies graph neural networks (GNNs) with physics-based vehicle simulations and open global data to produce high-precision forecasts of heavy-duty (i.e., buses and trucks) EV charging needs. Forecasts are generated at the vehicle level along routes and then aggregated to fleet- or corridor-level demand using probabilistic priors over vehicle attributes. We validate its performance through large-scale simulations (including ten international virtual corridor case studies) and real-world truck data from Finland, revealing a 500-fold computational speedup over conventional physics-based approaches at only a marginal (<br/>4%) accuracy trade-off. By identifying peak periods and locations of corridor demand for specified fleets, WeTRaC can effectively mitigate grid overload and accelerate the transition toward zero-emission transport.