TSUNAMI EVACUATION GUIDANCE USING REINFORCEMENT LEARNING ALGORITHM

Mas E.; Moya L.; Koshimura S. · World Conference on Earthquake Engineering proceedings (2021)

Resumen

In previous tsunami disasters it was reported that many residents used vehicles to evacuate from the coast. Consequently traffic congestions delayed evacuation times reducing survival chances in the area. Similarly, when evacuating as pedestrian, safe evacuations can be compromised by overcrowding conditions due to high population concentration, narrow streets and limited space for mobility. Thus, finding the best evacuation route considering less congestion and higher chances to reach safe haven is critical for the evacuation process. We envision a future of smart and autonomous evacuations in a post-information age, where smart cities become fully interconnected and people or vehicles can be intelligently guided in real time with precise information aiming for an optimum behavior and a safe evacuation. With this in mind, we built a model of evacuation with limited evacuee decision-making, however with learning-skilled connections within the network that would guide the flow of evacuees in directions that result best to reduce congestion and increase the chances to survive. One can imagine also smart and interconnected digital tsunami sign boards placed at intersections guiding evacuees accordingly to current congestion conditions in the network. Here we used reinforcement learning to find the best policy (routing) constrained to low density in roads and successful guidance behavior of nodes in the network. © The 17th World Conference on Earthquake Engineering.

Instituciones: Universidad Nacional de Ingeniería