Latent behavioural modelling of multi-dimensional vehicle parking patterns for urban EV charging planning

Understanding vehicle parking behaviour in metropolitan environments is essential for the effective deployment of electric vehicle charging infrastructure. This study introduces a data-driven framework that integrates spatial clustering with probabilistic topic modelling to characterise urban stop patterns in a dense, diverse urban context. Using large-scale GPS-based vehicle trajectory data across Greater London, the research first identifies meaningful parking areas via Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN), which can detect high-density stop regions shaped by complex land-use structures. On this spatial foundation, a latent topic model is adapted to extract underlying behavioural patterns associated with individual vehicle stops. The model incorporates multiple contextual attributes, including arrival time, stop duration, location category, and day of week, enabling the identification of recurring activity types such as overnight residential stays, short-duration commercial visits, and medium-length service stops, with 3-, 5-, and 10-topic models progressively revealing behavioural complexity. Each activity type reflects a unique temporal and spatial signature. The resulting behavioural segmentation supports charging infrastructure decisions by aligning charger type and scheduling with observed usage profiles. The London case study demonstrates how such a framework enables targeted, adaptive, and context-sensitive deployment of charging assets in complex urban systems.