ダム管理のための機械学習を用いたハイブリッド型流入量予測モデルの検討

抄録

Recently, as it occurs flood disasters due to climate change, it is important to operate dams. Especially, for hydroelectrical power plants of the dams, it is also necessary to predict the discharge to the dams appropriately at the phase of during usual and flood water for managing power generation and preparing for flood. However, conventional methods to predict the discharge have been developed at each phase of usual and flood water separately. In this report, we developed a hybrid-discharge-prediction model, which is composed by a state discriminator and discharge prediction models with machine learning and a flood simulator. This hybrid-discharge-prediction model can detect the state of discharge and adopt an appropriate discharge prediction model each state of discharge and prediction time. As a result, it was shown that the hybrid-discharge-prediction model can detect 7 states and predict the discharge to the dam in 5 hours at the phase from usual to flood water.