Dynamic Evacuation Shelter Allocation in Response to Human Mobility: A Case Study of Taipei City

Natural disasters often occur unexpectedly, catching people off guard, such as the Hualien earthquake on 3 April 2024. Many of the evacuation-monitoring systems currently in place lack real-time updates of shelter capacities, which raises the risk of overcrowding under wartime scenarios. This study developed a system for the targeted assignment of evacuation sites during air raids. The DBSCAN algorithm was used to group data based on pedestrian flow patterns and an LSTM model was used to enhance the prediction speed and accuracy. Weighted Voronoi diagrams delineated regions to identify optimal evacuation points, while real-time SMS notifications through base station positioning disseminated evacuation information to the public. The experiment results demonstrated the effectiveness of the proposed system in facilitating safe evacuations across a broad range of geographic regions while reducing the number of LSTM models. Dynamic updates on the shelter capacities make it possible for citizens to make informed decisions during air raid emergencies.