[H-DS11] Human environment and disaster risk

[HDS11-P02] Tsunami evacuation behaviour optimisation and dynamic evacuation guidance in Kanazawa-Hakkei using multi-agent simulation

*Miho Fujita¹, Ahyi KIM¹ (1.Yokohama City University )

In Japan, there have been many large-scale earthquakes and the resulting disasters, especially tsunamis, have caused extensive damage. In the near future, there is a concern that a major earthquake could occur in the Tokyo metropolitan area. In the event of a major earthquake, coastal areas along the Pacific coast are expected to be inundated by a tsunami. In order to reduce the number of tsunami victims, it is necessary to ensure rapid and safe evacuation. In this study, we propose the installation of dynamic evacuation guidance signs in the Kanazawa-Hakkei area, Yokohama City, in order to improve the efficiency of tsunami evacuation behavior and verify the effectiveness of the proposed system. Furthermore, the optimal locations of the proposed signboards are obtained by NSGA-II using the recognition rate, recognition time, and the number of signs as objective functions. In this study, we confirm the effectiveness of the proposed billboards under the assumption that all signboards are in normal operation, and do not discuss the uncertainty of the system. The proposed signboard has a function to propose the best evacuation route in real time by using QR codes, taking into account the attributes of evacuees, the time remaining until the tsunami arrives, and the number of people who can be accommodated in evacuation shelters. When a route is proposed, a safe zone such as high ground is set as the first priority destination, but if it is judged to be unreachable based on the remaining time until the tsunami arrives, an evacuation shelter is set as the destination. In this case, the current capacity of the evacuation shelter is taken into account to avoid the situation in which the number of evacuees exceeds the capacity after the arrival of the tsunami. Specifically, when an agent is assigned to a shelter, we update the available capacity of the shelter so that shelter’s capacity is not exceeded. In addition, when selecting a shelter, the distance from the coast and the location of the shelter, such as its elevation, are taken into account, and the shelters are given a safety ranking, with the one receiving the highest ranking being given priority as a destination. The effectiveness of the proposed signboards is verified using a multi-agent simulation (hereinafter referred to as MAS). MAS is a method used to reproduce complex dynamic phenomena and to analyze their behaviors through interactions among multiple autonomous agents. In this simulation, we assume tsunami evacuation behavior after a Keicho-type earthquake, which has the highest predicted maximum tsunami height. Each agent is given two attributes, i.e., walking speed and evacuation start time, and is classified into three types of behavior: aiming for the nearest evacuation site, aiming for a safe zone such as high ground, and following others. Based on each attribute and movement type, agents select different evacuation routes and action patterns. The simulation results show that the success rate of evacuation is significantly improved by the installation of signboards. Furthermore, NSGA-II is executed and the solution with the minimum number of signboards is selected as the optimal solution. We then placed the signboards at the selected locations and conducted the simulation again. Although the success rate of evacuation was improved, some agents did not receive proper guidance and become lost in some cases. In the future, we plan to analyze the causes of agents who could not receive proper guidance, to verify the case in which the hyperparameters of NSGA-II are changed, and to further select the optimal placement locations.