As urban populations continue to grow, traffic congestion has intensified, leading to the construction of 3,809 tunnels in South Korea by 2023. The semi-enclosed structure of tunnels poses serious risks during fire incidents, such as toxic gas accumul...
As urban populations continue to grow, traffic congestion has intensified, leading to the construction of 3,809 tunnels in South Korea by 2023. The semi-enclosed structure of tunnels poses serious risks during fire incidents, such as toxic gas accumulation, heat buildup, and limited accessibility for firefighting. From 2019 to 2023, a total of 106 tunnel fire incidents were reported in Korea, including a major fire in the Samae Tunnel that resulted in significant casualties. The tunnel design process involves route selection, cross-sectional planning, ventilation system design, and iterative simulations to ensure safety under fire conditions. However, this process is time-consuming, and if the feasibility of the route selected during the basic planning stage is later found problematic, it necessitates regression to the planning phase, causing delays in the overall tunnel design process. To address this issue, this study developed a predictive model for tunnel ventilation design based on national datasets and design guidelines. Fire simulations were conducted to evaluate the influence of structural variables, fire intensity, and ventilation performance. For longitudinal ventilation (binary classification), logistic/probit regression and machine learning models were used. For transverse ventilation (continuous prediction), Tobit regression and machine learning techniques were applied. The machine learning models outperformed traditional statistical regression, with Extreme Gradient Boosting and Extra Trees identified as the most effective. External validation was conducted using real-world tunnel data. The model for longitudinal ventilation (Extreme Gradient Boosting) achieved 0.9697 accuracy, while the model for transverse ventilation (Extra Trees) demonstrated an explanatory power (R²) of 0.9061. These results confirm the practical applicability of the proposed models. The optimized formulas and predictive models developed in this study facilitate early-stage ventilation planning, reduce iterative design steps, and improve design efficiency. Additionally, the methodology can be applied to upgrade ventilation systems in existing tunnels and serve as foundational research for integrating future smart ventilation systems and IoT technologies.