Real-time RSET prediction across three types of geometries and simulation training dataset: A comparative study of machine learning models
Loading...
Date
2024-05-24
Authors
Uhlík, Ondřej
Okřinová, Petra
Tokarevskikh, Artem
Apeltauer, Tomáš
Apeltauer, Jiří
Advisor
Referee
Mark
Journal Title
Journal ISSN
Volume Title
Publisher
Elsevier
Altmetrics
Abstract
Agent-based evacuation models provide useful data of the evacuation process, but they are not primarily designed for use during an emergency. The paper aims to test predicting RSET using a surrogate ML model trained on a simulation dataset with 60 samples. A total of 9 machine learning algorithms were tested on 3 simple geometries: bottleneck, stairway and walkway. A set of 7 spatial features was used to train the surrogate models. The results showed a relatively good ability of Artificial Neural Network to learn in scenarios involving bottlenecks and stairways, with an R2: 0.99 on the testing dataset. In the walkway scenario, all models experienced a significant drop in performance, with Gradient Boost performing the best (R2: 0.92). The paper demonstrated ability to generalize effectively in bottleneck-type tasks with training on a relatively small dataset containing spatial parameters obtainable in real-time from camera systems.
Description
Citation
Developments in the Built Environment. 2024, vol. 18, issue 100461, 13 p.
https://www.sciencedirect.com/science/article/pii/S266616592400142X?via%3Dihub
https://www.sciencedirect.com/science/article/pii/S266616592400142X?via%3Dihub
Document type
Peer-reviewed
Document version
Published version
Date of access to the full text
Language of document
en