深層学習の適用によるニューラルネットワーク洪水予測の精度向上

抄録

In flood prediction, reduction of the uncertainty is one of the biggest issues. As a novel river stage prediction model, the artificial neural network model which is trained by the deep learning method was developed. The model is composed of 4 layer feed-forward network. As a network training method, stochastic gradient descent method based on the back propagation method was applied. Input of the model is hourly change of water level and hourly rainfall, output data is water level of prediction point. Developed model was applied to 4 rivers in Japan, OOYODO River, KOKAI River, ONGA River and KANO River.

As a result, there was a significant improvement at OOYODO and ONGA River, which has relatively many observation stations. In these basins, enhance of the learning ability may contributed to the result. On the other hand, at KOKAI and KANO River, there may be little margin for improvement even though learning ability enhanced.