Development of Disaster Prevention System Based on Deep Neural Network using Deep Learning with Dropout

Yoon, I.J. and Hong, J.-W., 2019. Development of disaster prevention system based on deep neural network using deep learning with dropout. In: Lee, J.L.; Yoon, J.-S.; Cho, W.C.; Muin, M., and Lee, J. (eds.), The 3rd International Water Safety Symposium. Journal of Coastal Research, Special Issue No. 91, pp. 186-190. Coconut Creek (Florida), ISSN 0749-0208. Recently, unprecedented abnormal climates and disasters of the greatest intensity have increased worldwide due to various problems such as global warming. In particular, from the Great East Japan Eartquake in March 2011, Japan has suffered from the highest frequency of disasters (floods, earthquakes, etc.) and the single largest external force (tsunami) that have not been experienced before. For such disasters of the maximum frequency, the hard and soft countermeasures alone have limitations and it is critical to integrate and appropriately harmonize them with various disaster prevention systems. The purpose of this study is to develop a flood risk management information system to support natural disaster mitigation measures. The main function of the disaster prevention support system is to select the most similar one to a typhoon before landing based on past typhoon data and artificial intelligence (AI) technology and to present guidelines for prior disaster prevention activities according to various disaster factors caused by the selected typhoon (e.g., water level and river flow rate). A database was constructed for 189 typhoons that directly or indirectly affected the target area among those that occurred between 1950 and 2017, and a system was developed by applying AI technologies such as Deep Neural Network (DNN) and Dropout. The main parameters of this database are typhoon route, central pressure, moving speed, precipitation, water level, and flow rate. For the grid sizes of the model, 2° and 1° in longitude and in latitude were considered. The system search results showed the highest reliability of the model at the grid size of 1° when only typhoon route, center pressure, and moving speed were considered compared to the results of considering other parameters.