Synthetic time series generation for reproducing driver behavior in Electric Vehicles

The rapid adoption of electric vehicles (EVs) is transforming transportation and energy systems, creating new opportunities for data-based optimization of battery use, charging infrastructure, and energy management. However, the limited accessibility of real-world EV data due to privacy constraints, proprietary restrictions, and high collection costs poses significant challenges for developing and validating analytical and predictive models. To address this limitation, this thesis proposes a comprehensive framework for generating and evaluating synthetic electric vehicle time-series data that reproduces the temporal, statistical, and behavioral characteristics of real-world operation. The approach models two fundamental event types, trips and charging sessions, which form irregular, event-driven sequences with strong causal dependencies. To capture these dynamics, we propose an adaptation of the KoVAE generative model to produce short temporal windows of synthetic data that preserve local dependencies between trips and charging events. We then apply a domain-informed post-processing pipeline that enforces physical and heuristic constraints to ensure plausibility and behavioral realism. Finally, we reconstruct validated windows into coherent long-term timelines through an alignment process that maintains temporal and energetic consistency and driving-profile coherence. The quality of the generated data is assessed through a multi-level evaluation framework encompassing both quantitative and qualitative analyses. Standard methods are employed to assess the fidelity, diversity, and usefulness of the generated data. Additionally, expert-based and heuristic evaluations are applied to reconstructed timelines to verify behavioral faithfulness, physical validity, and long-term consistency of battery state-of-charge trajectories. Overall, this work presents an integrated end-to-end methodology for realistic synthetic generation of EV time-series data, enabling experimentation in mobility and energy contexts where real-world data are scarce or inaccessible.