Diffusion Models Enhance EV Fleet Energy Prediction for Efficient Routing

Commercial delivery fleets are increasingly electrifying, shifting routing optimization from simple distance and time to energy efficiency. Current models often provide only deterministic estimates, lacking the full range of plausible energy consumption trajectories needed for robust operational decisions. This new framework addresses this by using conditional diffusion models. The proposed model generates realistic EV battery-current profiles, taking into account factors like vehicle speed and ambient temperature. It combines a latent conditioning encoder with a temporal 1D U-Net denoising backbone, allowing trip-related conditions to guide the generation process. Evaluated on a large commercial EV telemetry dataset, the model successfully generates current trajectories that capture both overall trends and sharp transient events. It significantly outperforms direct condition injection methods, demonstrating a robust generative modeling approach for understanding EV energy consumption under real-world conditions, which is crucial for large-scale fleet planning.