Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Real-time prediction of tunnel evacuation time using machine-learning surrogate models

Uhlík, Ondřej ; Gonzales-Villa, Javier ; Cuesta, Arturo ; Ronchi, Enrico LU orcid ; Juřík, Vojtěch ; Juránková, Radka ; Apeltauer, Tomáš and Apeltauer, Jiří (2026) In Safety Science 203.
Abstract

While advanced microscopic models provide detailed behavioural representation, their computational cost remains a significant barrier to real-time operational use. This study explores the use of machine-learning surrogate models for real-time estimation of tunnel required safe egress time. A synthetic dataset of 50,910 evacuation simulations was created using procedurally generated simulations with variable tunnel geometry and pseudo-random sampling using the Pathfinder evacuation simulator. Two complementary surrogate model classes were investigated: a tree-based extreme gradient boosting ensemble and a multilayer perceptron. Models were trained using nested cross-validation and validated against 10 full-scale tunnel evacuation... (More)

While advanced microscopic models provide detailed behavioural representation, their computational cost remains a significant barrier to real-time operational use. This study explores the use of machine-learning surrogate models for real-time estimation of tunnel required safe egress time. A synthetic dataset of 50,910 evacuation simulations was created using procedurally generated simulations with variable tunnel geometry and pseudo-random sampling using the Pathfinder evacuation simulator. Two complementary surrogate model classes were investigated: a tree-based extreme gradient boosting ensemble and a multilayer perceptron. Models were trained using nested cross-validation and validated against 10 full-scale tunnel evacuation experiments conducted in three countries. The extreme gradient boosting model demonstrated stable generalisation performance, low prediction error and consistent conservative bias across all validation datasets (R2: 0.99, MAE: 9.2 s). The multilayer perceptron showed reduced robustness under experimental conditions (R2: 0.91, MAE: 32.1 s). Results indicate that tree-based surrogate models are able to mimic required safe egress time predictions performed with more computationally expensive agent-based simulators. The proposed approach enables rapid evacuation-time assessment suitable for operational tunnel emergency decision support.

(Less)
Please use this url to cite or link to this publication:
author
; ; ; ; ; ; and
organization
publishing date
type
Contribution to journal
publication status
published
subject
keywords
Agent-based models, Evacuation, Machine learning, Real-time assessment, Road tunnels
in
Safety Science
volume
203
article number
107348
publisher
Elsevier
external identifiers
  • scopus:105043040814
ISSN
0925-7535
DOI
10.1016/j.ssci.2026.107348
language
English
LU publication?
yes
id
4549278b-79a7-4ee9-8ba2-5a0c5cf5438c
date added to LUP
2026-09-30 13:03:28
date last changed
2026-09-30 13:03:55
@article{4549278b-79a7-4ee9-8ba2-5a0c5cf5438c,
  abstract     = {{<p>While advanced microscopic models provide detailed behavioural representation, their computational cost remains a significant barrier to real-time operational use. This study explores the use of machine-learning surrogate models for real-time estimation of tunnel required safe egress time. A synthetic dataset of 50,910 evacuation simulations was created using procedurally generated simulations with variable tunnel geometry and pseudo-random sampling using the Pathfinder evacuation simulator. Two complementary surrogate model classes were investigated: a tree-based extreme gradient boosting ensemble and a multilayer perceptron. Models were trained using nested cross-validation and validated against 10 full-scale tunnel evacuation experiments conducted in three countries. The extreme gradient boosting model demonstrated stable generalisation performance, low prediction error and consistent conservative bias across all validation datasets (R<sup>2</sup>: 0.99, MAE: 9.2 s). The multilayer perceptron showed reduced robustness under experimental conditions (R<sup>2</sup>: 0.91, MAE: 32.1 s). Results indicate that tree-based surrogate models are able to mimic required safe egress time predictions performed with more computationally expensive agent-based simulators. The proposed approach enables rapid evacuation-time assessment suitable for operational tunnel emergency decision support.</p>}},
  author       = {{Uhlík, Ondřej and Gonzales-Villa, Javier and Cuesta, Arturo and Ronchi, Enrico and Juřík, Vojtěch and Juránková, Radka and Apeltauer, Tomáš and Apeltauer, Jiří}},
  issn         = {{0925-7535}},
  keywords     = {{Agent-based models; Evacuation; Machine learning; Real-time assessment; Road tunnels}},
  language     = {{eng}},
  publisher    = {{Elsevier}},
  series       = {{Safety Science}},
  title        = {{Real-time prediction of tunnel evacuation time using machine-learning surrogate models}},
  url          = {{http://dx.doi.org/10.1016/j.ssci.2026.107348}},
  doi          = {{10.1016/j.ssci.2026.107348}},
  volume       = {{203}},
  year         = {{2026}},
}