Machine learning for fire evacuation assessment in road tunnel fires

Machine learning for fire evacuation assessment in road tunnel fires

Stať ve sborníku mimo WoS a Scopus

Machine Learning (ML) is a promising tool to perform evacuation predictions in relatively short time. Nevertheless, to date, many algorithms exist and present different strengths and weaknesses in terms of key features such as their complexity and ability to predict given outcomes. This study investigates the use of ML to predict human consequences of road tunnel fires. Given the exploratory nature of this work, nine algorithms were trained on synthetic stochastic datasets to develop the predictive models. Key outputs adopted to evaluate the model performance included total evacuation times and number of people affected by fire. A feature removal process was also conducted to inform the development of efficient, interpretable, generalizable, and accurate predictions for supporting fire risk assessments in road tunnels.

Machine Learning (ML) is a promising tool to perform evacuation predictions in relatively short time. Nevertheless, to date, many algorithms exist and present different strengths and weaknesses in terms of key features such as their complexity and ability to predict given outcomes. This study investigates the use of ML to predict human consequences of road tunnel fires. Given the exploratory nature of this work, nine algorithms were trained on synthetic stochastic datasets to develop the predictive models. Key outputs adopted to evaluate the model performance included total evacuation times and number of people affected by fire. A feature removal process was also conducted to inform the development of efficient, interpretable, generalizable, and accurate predictions for supporting fire risk assessments in road tunnels.

machine learning, evacuation modeling, tunnel, safety

machine learning, evacuation modeling, tunnel, safety

CUESTA, A.; RONCHI, E.; UHLÍK, O.; GONZALES-VILLA, J.; ALVEAR, D.

30.06.2025

11