Author(s): Mengtong Li; Tokiyosh Hosono; Tomoharu Hori
Linked Author(s): Tomoharu Hori
Keywords: Flood disasters; Intelligent Flood Evacuation (IFE); Deep learning; Scoring function
Abstract: Efficient evacuation is quite important for decreasing the casualties in flood disasters. Flood hazard map is helpful for residents to recognize the necessity of evacuation but not sufficient for making the criteria to determine when to evacuate, where to go and which path to select dynamically according to the situations during flooding. Learning from experience is important to make evacuation action criteria such as critical water levels in nearby rivers to start evacuation, accumulated rainfall amount to start evacuation, pathway to the shelters according to the flooding condition and so on. Human lifespan, however, is not so long compared to the return period of extreme floods that people have not so many chances to experience flood events. In order to overcome the lack of experience of residents and to make an effective evacuation framework, an Intelligent Flood Evacuation (IFE) model is developed in this study. Based on the evacuation simulations for various scales of flood events (simulated flood scenarios), the scoring function is derived to give how appropriate the time is to start evacuation. The time is judged most appropriate when the agent encounters flooding in the moving process to shelters if it starts evacuation just after this time. The function gives low scores if the agent starts after or much earlier than this timing. Then IFE model learns the timing to start evacuation with reference to the timing scoring and flood scenarios: it builds the artificial neural network to derive the appropriate timing to start evacuation based on the water levels of neighboring rivers, accumulated rainfall and so on. The IFE employs simple back propagation network and deep learning network according to the size of inputs and outputs. Besides the timing of evacuation, the possibility to learn the selection of the destination shelter and pathway will be also discussed.
Year: 2020