A hybrid agent-based and Bayesian inference framework for spatiotemporal occupancy modeling in metro stations without real-time origin-destination data

Accurately modeling the spatiotemporal occupancy in metro stations is essential for energy management and improving indoor environmental quality. However, traditional occupancy movement models depend on a known passenger start and an endpoint in the station to reconstruct routes, which metro stations generally lack, thereby posing challenges for distribution modeling. Existing metro station occupancy models rely heavily on complete network-wide origin-destination data, which is often unavailable or delayed at the station operational level, or sensor-based indoor tracking. To address these challenges, this study introduces a hybrid agent-based and Bayesian inference framework to estimate metro station occupancy distributions using only real-time automated fare collection (AFC) data. Passenger dwell time distributions are updated using Hamiltonian Monte Carlo (HMC) sampling. The model achieves high accuracy with R² values of 0.99 and a MAPE of 11.5%. Furthermore, the simulation shows that the dynamic occupancy-based HVAC system achieves 30.21% energy reduction when compared to the fixed-occupancy schedule. This approach enables scalable, data-efficient estimation of in-station passenger dynamics, offering practical insights for operations, ventilation control, and infrastructure planning.