Study on the Mitigation of Evacuation Bottlenecks in Deep Underground Tunnels through Deep Learning-Integrated Simulation

In this study, potential bottleneck phenomena in the evacuation connection passage at the Shingu vertical shaft of the Seobu Underground Expressway were quantitatively analyzed using evacuation simulations, and alternative structural design options were compared. The existing configuration, which relies on a single exit and has geometric constraints at curved sections, resulted in an average evacuation time of 739.5 s and a maximum evacuation time of 1,414.2 s, with significant temporal variability in evacuation flow. To avoid large-scale structural modifications such as cross-sectional expansion, improvement scenarios were developed by sequentially applying less intrusive measures, including multiple exit doors, passage straightening, and buffer spaces (hereafter referred to as pockets). The scenario incorporating pockets achieved reductions of 41.8% and 48.0% in average and maximum evacuation times, respectively, while also mitigating fluctuations in exit flow. Although these findings are limited to a specific case setting and simulation conditions, this study provides fundamental insights into the timing of bottleneck formation and changes in flow stability within deep evacuation connection passages. These results can serve as a reference for performance-based design to evaluate and compare evacuation design alternatives.