Prediction of EV Charging Load Based on User Equilibrium and MPNN

This paper proposes a physically consistent framework for predicting electric-vehicle (EV) charging loads by integrating user-equilibrium (UE) traffic assignment with a message passing neural network (MPNN). The UE model with feasible-path generation produces station-level load labels, while the MPNN with station masking predicts loads directly from transportation network graphs and their node and edge features. Node attributes encode station identity and origin–destination (OD) demand, and edge attributes include link capacity, free-flow travel time, and link length. Using a network with 20 nodes and 8 charging stations, we generate 5000 scenarios for training and evaluation. Results demonstrate that the proposed method achieves near-UE accuracy while reducing per-scenario computation from seconds to milliseconds. This balance of accuracy and efficiency makes the approach well suited for rapid analysis and real-time online deployment.