DAM INFLOW PREDICTION BY DEEP LEARNING USING RADAR RAINFALL

Abstract

We developed the deep learning model for dam inflow prediction. The input data of the model was radar rainfall. We applied the model to the Shimouke Dam and confirmed the good result up to 60 minutes prediction. We compared the deep learning model with conventional 3-layer neural network, and confirmed the superiority of the deep learning. We also compared the developed model with the deep learning model trained with rain-gauge data, and confirmed the superiority of the radar rainfall. In the end, we applied the prediction rainfall data to the input data of the model. We confirmed that there was no effect of the rainfall prediction error up to 20 minutes prediction in this model. In 60 minutes prediction, effect of the rainfall error appeared especially in the rising limb of the flood.