Graph neural networks for building evacuation prediction : integrating real-time pedestrian flow simulation into BIM workflows

Tools

Berggold, Patrick and Borrmann, André and Nousias, Stavros; Moreno-Rangel, Alejandro and Kumar, Bimal, eds. (2025) Graph neural networks for building evacuation prediction : integrating real-time pedestrian flow simulation into BIM workflows. In: EG-ICE 2025. University of Strathclyde Publishing, GBR. ISBN 9781914241826 (https://doi.org/10.17868/strath.00093235)

[thumbnail of Berggold-etal-EG-ICE-2025-Graph-neural-networks-for-building-evacuation-prediction]
Preview
 Text. Filename: Berggold-etal-EG-ICE-2025-Graph-neural-networks-for-building-evacuation-prediction.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo
Download (1MB)| Preview

Abstract

Effective evacuation planning requires accurate pedestrian simulations, yet integrating such models into early-stage architectural design remains challenging due to manual data conversions and long simulation runtimes. Meanwhile, Building Information Modelling (BIM) is widely used for digital building representation but lacks native pedestrian simulation support. To address this gap, we propose a novel approach that combines graph-based building representations with a Graph Neural Network to predict pedestrian flow in real-time. Our method transforms parametric BIM layouts into graph representations, enabling fine-grained temporal resolution of crowd dynamics. Unlike previous surrogate models with static or coarse outputs, our approach captures density rates dynamically, facilitating rapid safety assessments during design iterations. This integration streamlines pedestrian simulation within BIM workflows, allowing for layout optimization layouts to ensure safe and efficient evacuation design.

CORE (COnnecting REpositories)