Multi‑level driving behavior modeling and hydrogen refueling analysis in fuel cell vehicles using big data

• Developed a multi-dimensional driver profile for FCEVs including both driving and stationary behavior for FCEVs using high-resolution big datasets • Developed a PCA–UMAP–KMeans pipeline for describing driving behaviors effectively in driving styles • Developed a clustering pipeline for describing stationary behaviors especially in hydrogen refueling characteristics This study proposes a new data-driven framework for analyzing driver behavior using high-frequency data from fuel cell vehicles (FCEVs). The proposed pipeline involves data collection and processing, trip segmentation, segment-level preprocessing, segment-level features extraction and processing, and unsupervised clustering across both dynamic and stationary behavior. Utilizing a PCA–UMAP–KMeans pipeline, we successfully classified driving styles into three distinct categories: aggressive, moderate, and conservative. Stationary behavior analysis, integrated with location data, identifies hydrogen refueling as a key operational feature. Quantitative findings reveal a significant refueling peak between 14:00 and 16:00, with typical refill durations highly concentrated between 3 and 15 minutes, mimicking conventional vehicle convenience. Furthermore, correlation analysis demonstrates that aggressive drivers exhibit higher equivalent hydrogen consumption rates, leading to increased refueling frequencies and premature refueling habits at higher remaining fuel levels. Finally, dynamic driving styles and stationary patterns are integrated into individual driver profiles, providing a quantitative foundation for adaptive energy management and durability assessment in FCEVs.