A CASE STUDY OF FLOOD WATER LEVEL PREDICTION IN THE TOKORO RIVER IN 2016 USING RECURRENT NEURAL NETWORKS

DOI Web Site 3 Citations 7 References

Bibliographic Information

Other Title
  • リカレントニューラルネットワークを用いた2016年の常呂川洪水事例の水位予測

Abstract

<p> Recently, machine learning technologies such as deep neural networks have increasingly applied for floodwater-level prediction. However, in many of these studies, water level time series data were modeled inappropriate. In this paper, we propose a flood prediction method using recurrent neural networks (RNNs) that can process time series data. We evaluated water level prediction model using the RNNs model to predict the Tokoro River’s water level during the heavy rainfall disaster in Hokkaido in August 2016. Moreover, we generated a multi-node output model that simultaneously predicts the water level at five points. For performance evaluation, we compared conventional fully connected neural networks (FCNNs) and RNNs using the root mean square error (RMSE) between the measured and predicted water levels. With a lead time of 6 hour, the RMSE of the RNNs model and FCNNs model are 0.29 m and 0.40 m, respectively. The above results demonstrate the effectiveness of flood prediction using an RNNs with a time series structure.</p>

Journal

Citations (3)*help

See more

References(7)*help

See more

Keywords

Details 詳細情報について

Report a problem

Back to top