Cabras, StefanoHe, Sun2024-05-272024-05-272024-012024-01-19https://hdl.handle.net/10016/43907The thesis defended here is that modeling passenger flows with acceptable properties can be done with the models exposed in the above chapters. This thesis explored statistical methods for modeling passenger flows in Beijing’s Metro. The focus was on Bayesian methods and their application to dynamic systems, particularly urban metros. The Bayesian paradigm, including prior probability, likelihood, and posterior probability, was emphasized. Computational challenges were addressed using Integrated Nested Laplace Approximations (INLA), suitable for large-dimensional parameters in complex models. The work started from the socio-economic and engineering contexts of Beijing’s rapid urbanization. The Beijing Metro, serving millions daily, faces challenges due to unpredictable ridership patterns influenced by various factors. Predictive analytics are therefore crucial for operational efficiency, expansion planning, and passenger experience enhancement. Bayesian analysis was used for its adaptability and learning capability from new data. INLA was employed for efficient Bayesian inference, particularly in complex spatial and spatio-temporal models relevant to the study. The framework proved effective for regression models, dynamic linear models, and spatial applications. Data from ticketing systems, turnstiles, smart card check-ins, and mobile apps provided essential input for analysis. This data was crucial for our research as it is for managing peak traffic, scheduling trains, ensuring passenger safety, and supporting strategic decision-making. The thesis demonstrated the effectiveness of Bayesian models in predicting passenger flow in urban metro systems. Future work could focus on enhancing the computational efficiency of these models and exploring their application in other dynamic urban systems. Further research could also delve into the integration of additional data sources and the development of more advanced predictive models. Although the primary focus is on the Beijing Metro, this research draws data from the 1st of September to the 31st of October in 2020 to ensure a comprehensive understanding. It is worth noting that while the Bayesian model developed might offer theoretical applications for other metro systems, its design, calibration, and validation remain rooted in Beijing’s context. Aspects like intermodal transportation or predictions for bus networks fall outside of this study’s purview. To our knowledge, at the moment the daily passenger model has also been fitted to data from the metro network in other cities with a performance similar to the one shown here.engAttribution-NonCommercial-NoDerivatives 4.0 InternationalBayesian model callibrationIntegrated nested Laplace approximationSpatial–temporal modellingPoisson countsProportional odds modelDeep learningMetro transportationBeijingdoctoral thesisEstadísticaopen access