Synthetic data generation for joint electric vehicle driving and charging events via deep generative networks

Abstract

The rapid growth of the global electric vehicle (EV) market has generated vast amounts of real-time data through smart vehicle devices, capturing detailed EV behavioral patterns, including spatiotemporal charging distribution, electricity consumption, origin-destination (OD) driving trajectories, and parking dynamics. This data offers transformative potential for refining transportation policy, strategically deploying charging infrastructure, and facilitating vehicle-to-grid (V2G) energy integration.

Recent advances in generative modeling offer solutions by creating synthetic datasets that maintain data utility while protecting privacy. However, existing studies often lack detailed joint generation of driving and charging events and are less effective in handling highly dependent and constrained event variables. To address these gaps, we propose a joint EV driving and charging model for generating large-scale spatiotemporal behavior data by combining Transformer architecture and Gibbs sampling, leveraging 15 months of data from 3777 EVs in Shanghai, China. Our model effectively captures inter-feature and temporal dependencies, enabling scenario-based data generation while addressing privacy challenges. Our codebase, pre-trained model, and synthetic dataset are publicly available at: https://github.com/zhilee2023/EV-drive-charge-data-gen