REAL-TIME FLOOD FORECAST USING THE RRI MODEL AND A DEEP NEURAL NETWORK

Abstract

 The main objective of this study is to implement the inexpensive and simple flood-forecast targeting for a mountainous river. We combined the RRI model with a Deep Neural Network (DNN), a type of artificial intelligence technology. A hierarchical neural network is applied to DNN, and the hidden layer is composed by the multiple layers. In this study, the hidden layer was set to three layers, and twenty neurons in each layer. The method of learning for DNN was tested the batch-learning with past floods supervised and the online-machine-learning that updates learning data with current information. The batch-learning was The reproducibility of DNN by the batch-learning was higher than that of the H-Q method. However, a bias error was confirmed in the low water level. In DNN by the online-machine-learning, it was confirmed that high accuracy and stability prediction can be made by learning data higher than the flooding caution water level.