FloGAN: Scenario-Based Urban Mobility Flow Generation via Conditional GANs and Dynamic Region Decoupling
Seanglidet Yean 1 , Jiazu Zhou 1 , Bu-Sung Lee 2 , Markus Schläpfer 3 · 2025
Read FloGAN: Scenario-Based Urban Mobility Flow Generation via Conditional GANs and Dynamic Region Decoupling by Seanglidet Yean 1 , Jiazu Zhou 1 , Bu-Sung Lee 2 , Markus Schläpfer 3 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-12053.txt.
Introduction
City planners need to test how people will move as land use and population change, but past travel records cannot describe a future district. Seanglidet Yean and co-authors present FloGAN, a method that generates origin-destination travel flows for made-up urban scenarios. It uses conditional generative adversarial networks, fed with historical data plus flexible factors such as shifting region sizes and land use types, with regions decoupled dynamically as scenarios change.
The method produces flows quickly, at whatever spatial grain the planner wants, and needs neither heavy calibration data nor detailed behaviour models. The team demonstrates it on mobile phone data from Singapore and compares it with existing approaches, reporting promising performance. FloGAN aims squarely at forward-looking questions — new towns, rezoned districts, changed populations — where history alone cannot guide transport planning or sustainable city design.
Catalog record
| CREATOR | Seanglidet Yean 1 , Jiazu Zhou 1 , Bu-Sung Lee 2 , Markus Schläpfer 3 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-12053
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
FloGAN: Scenario-Based Urban Mobility Flow Generation via Conditional GANs and Dynamic Region Decoupling Seanglidet Yean 1 , Jiazu Zhou 1 , Bu-Sung Lee 2 , Markus Schläpfer 3 Abstract The mobility patterns of people in cities evolve alongside changes in land use and population. This makes it crucial for urban planners to simulate and analyze human mobility patterns for purposes such as transportation optimization and sustainable urban development. Existing generative models borrowed from machine learning rely heavily on historical trajectories and often overlook evolving factors like changes in population density and land use. Mechanistic approaches incorporate population density and facility distribution but assume static scenarios, limiting their utility for future projections where historical data for calibration is unavailable. This study introduces a novel, data-driven approach for generating origin-destination mobility flows tailored to simulated urban scenarios. Our method leverages adaptive factors such as dynamic region sizes and land use archetypes, and it utilizes conditional generative adversarial networks (cGANs) to blend historical data with these adaptive parameters. The approach facilitates rapid mobility flow generation with adjustable spatial granularity based on regions of interest, without requiring extensive calibration data or complex behavior modeling. The promising performance of our approach is demonstrated by its application to mobile phone data from Singapore, and by its comparison with existing methods. I Introduction As land use and population densities evolve, urban mobility patterns undergo significant transformation. For effective urban infrastructure planning, ranging from transportation network design [ 1 ] and traffic flow optimization [ 2 ] to sustainable urban development [ 3 ] , understanding both current mobility patterns and future scenarios is essential. Analyzing projected mobility flows enables planners to anticipate future urban dynamics and assess the impact of proposed interventions. Existing methods for mobility flow generation or simulation have primarily relied on models such as the gravity model [ 4 ] , the radiation model [ 5 ] , a recently introduced visitation law [ 6 ] , or the exploration and preferential return (EPR) model [ 7 ] , differing in their assumptions about human behavior and scale. While the gravity model, as well as the ‘parameter-free’ radiation and visitation models, focus on aggregate factors like distance and population, the EPR (and its extensions such as the TimeGeo framework [ 8 ] ) incorporate individual decision-making based on preferences; i.e. the desire to revisit or explore new locations. These methods require either parameter calibrations on historic data (gravity and EPR-based models), which limits their application for future scenarios, they may have limited predictive capabilities at fine-grained intra-urban scales (radiation model), or may not sufficiently consider location-specific factors such as accessibility (visitation law). Recently, reinforcement learning (RL) and generative adversarial networks (GANs) have emerged as promising techniques for mobility simulation, offering enhanced flexibility and predictive accuracy. RL has been used to model decision-making for mobility generation, where it models intelligent agent behavior and optimal paths in complex urban systems [ 9 ] . However, it highlights the high computational demands and challenges in producing reliable, realistic data, which is especially important when data is sparse or highly variable. Among these, GANs hold particular promise, as they can generate diverse, high-quality synthetic mobility data with less reliance on extensive real-world interaction and computational resources. In this paper, we explore the application of GANs for future mobility projections, aiming to enhance the reliability and scalability of urban mobility modeling and simulation. In the era of big data and data-driven methodologies, generative adversarial networks (GANs) have been utilised to simulate complex systems, including urban mobility, by generating synthetic data that closely mimics real-world patterns. In the context of mobility generation, existing methods like CollaGAN [ 10 ] have successfully integrated multiple data sources such as Household Travel Survey (HTS) and Transit Smart Card (SC) data to generate realistic activity schedules. However, these approaches are highly dependent on large, complex datasets, which can be difficult to obtain in data-scarce environments. In addition, other existing methods like MoGAN [ 11 ] and TrajGDM [ 12 ] , which focus on historical mobility trajectories, often struggle to account for changing factors such as land use and population density shifts. To address these challenges, this study proposes a novel GAN-based approach towards generating projected, scenario-based origin-destination (OD) mobility flows for a simulated urban environment. We intend to challenge the conventional application of GANs, which have seen great success in tasks such as RGB image generation with style changes, by applying them to mobility and OD flow generation. Our proposed method leverages GANs to fuse historical mobility data, such as GPS-based mobile phone data, with style-controlling options, such as base-map configurations, allowing for scalable and adaptable mobility simulations. This approach provides a solution for environments where multi-source, extensive data is scarce or unavailable, which enables accurate prediction of future urban dynamics and evaluation of proposed interventions without relying on a large amount of historical data. The key contributions of this paper are: • A data-driven approach to mobility flow generation that (1) requires a relatively small amount of calibration data, (2) operates independently of intricate behavior models requiring expert input, and (3) enables rapid data generation within seconds using a pre-trained model. • A flow decoupling technique leveraging dynamic mapping to enhance the flexibility of GANs for mobility simulations across various regions or specific areas of interest. • Enhanced user control over the generated data, ensuring adaptability to diverse planning and simulation needs. II Methodology Fig. 1 : Overview of the proposed FloGAN architecture. II-A FloGANs: Architecture Design To achieve the mobility OD flow generation with added conditions, we design a methodology, termed ‘FloGAN’ , consisting of three main components (Fig. 1 ): • Encoder/Decoder : This component performs representation transformation by converting individual-based trajectory data (GPS traces from mobile phones) into aggregated and structured OD flows. The transformation leverages a dynamic map representation as the spatial basis. The resulting OD flows are subsequently encoded as adjacency matrices to enable further analysis and learning. This approach overcomes the limitations that i ) GANs typically require a static input size (often requiring a 64 \times 64 image), and ii ) that an image is usually split into equally sized grid cells, which causes a loss in flexibility for specific region-to-region mobility flows. • Conditional Encoding : This component introduces auxiliary information to guide the generative process, enabling control over the ‘style’ or characteristics of the generated OD flow data (e.g., time of day, land use scenario). • Generative Model : This component represents the chosen GAN architecture. II-B Encoder/Decoder: Representation Transformation with Dynamic Region-Decoupling To address the trade-off between spatial resolution and computational efficiency, we introduce a dynamic map (see Fig. 2 ) that adaptively defines focus regions. The goal is to capture fine-grained mobility patterns in specific areas of interest while maintaining a coarser resolution elsewhere. This approach avoids training the model on the entire map at full resolution, which would be inefficient and prone to overfitting. Unlike static gridding approaches, the dynamic map enables selective spatial attention without violating the continuity of mobility flows across regions. Since OD flows inherently involve connections between distant zones, naive spatial partitioning is unsuitable. The dynamic map offers a solution by enabling localized focus while maintaining