Integrated Traffic–Weather-Aware Forecasting of Urban EV Charging Demand for Infrastructure Planning

Abstract

The accelerating adoption of electric vehicles (EVs) presents significant challenges for maintaining grid stability and optimizing charging infrastructure. Accurate short-term forecasting of EV charging demand is therefore critical to support reliable grid operation and effective energy management in urban environments. However, existing forecasting models often fail to capture the intricate interdependencies among mobility patterns, weather variations, and real-world charging behaviors, which constrains their generalizability and robustness. This study develops a multi-model forecasting framework that leverages Transformer-based deep learning architectures to integrate real-world charging data with traffic flow and meteorological variables for predicting short-term EV charging demand across metropolitan areas. To benchmark performance, two additional machine learning models—CatBoost and convolutional neural networks (CNNs)—are systematically evaluated using datasets from urban EV supply equipment (EVSE) and electric bus systems. The results indicate that Transformer-based models deliver superior predictive accuracy, temporal consistency, and adaptability compared with CNNs and CatBoost. Furthermore, sensitivity analysis reveals that traffic dynamics and user charging behavior exert the strongest influence on forecast performance. The proposed framework offers actionable insights for utilities and urban planners, facilitating resilient grid operation, optimized charging infrastructure deployment, and accelerated integration of EVs into the power system.