Generating high-resolution synthetic populations for transportation simulation--Open data applications for the BRUTUS modelling platform
Loading...
URL
Journal Title
Journal ISSN
Volume Title
School of Engineering |
Master's thesis
Authors
Date
Department
Major/Subject
Mcode
Degree programme
Language
en
Pages
74
Series
Abstract
The generation of high-resolution synthetic populations plays a crucial role in accurately modeling urban dynamics and travel demand, as it enables realistic representation of demographic characteristics and spatial heterogeneity. However, this process is increasingly constrained by the reduced accessibility to fine-grained official datasets due to confidentiality rules, rising costs, and access restrictions. This study proposes a three-stage framework to improve the demographic and spatial realism of synthetic populations for activity-based transportation models such as Ramboll’s BRUTUS, demonstrated in a case study of an island in Utrecht, the Netherlands. In the first stage, synthetic populations are generated from travel survey and census data using iterative proportional fitting (IPF), conditional tabular generative adversarial networks (CTGAN), and a hybrid approach, evaluated for their ability to reproduce marginal distributions, realistic attribute correlations, and other relevant metrics. We developed a multi-stage pipeline for synthetic population generation. In the second stage, synthetic households are formed from individual records, establishing demographic units prior to spatial allocation. A synthetic built-environment layer at the building level is then constructed using globally available open geospatial datasets, capturing both geometric and functional building attributes. In addition, we analyze the impact of each step in the pipeline on the final results, providing insight from the perspective of potential future users. In the third stage, households are assigned to specific buildings using a capacity-constrained method, ensuring alignment with census data. The results provide practical reference points for similar applications, particularly in contexts with limited official data, and highlight the potential of the framework not only for land use–transport integration and active mobility planning, but also for other domains where agent-based modeling is widely used.Description
Supervisor
Mladenovic, MilosThesis advisor
Hollestelle, MartijnJostmann, Jonas