APPLICATION OF DEEP LEARNING TO LONG-TERM PREDICTION OF RIVER STAGE

TOWARD GENERALIZATION OF LEARNING AND VERIFICATION PROCESS

Abstract

In this paper has (The authors have) conducted reviews of deep-learning models for the purpose of long-term forecasting necessary for the practical implementation of dams and river management. However, problems arose such as the process of setting the input and hyperparameters, as well as the reproducibility of floods for irregular pattern. The purpose of this study was to predict a river’s water level by the generalization process of setting the input and the hyperparameter in deep-learning model. Furthermore, the necessity of cross-validation for accuracy improvement and the effectiveness of iterative learning were examined. As a result, reproducibility was improved by adding the water level of the current time point at the input data. In addition, the applicability of floods with various waveforms and floods with less frequent occurrences were advanced by cross-validation with iterative learning.