Dam Inflow Prediction by Introducing Ensemble Rainfall Prediction into Machine Learning Methods

Abstract

This study proposes a dam inflow forecasting method that uses ensemble prediction for precipitation to reflect prediction uncertainty. The need to improve the accuracy of dam inflow prediction for effective dam operations has increased owing to large floods that have been occurring frequently across Japan in recent years. This study focuses on a dam in Hokkaido, Japan, which has been prone to floods in recent years. Elastic Net, which is a sparse modeling method, was used to predict inflows. Meso-scale Ensemble Prediction System (MEPS), the ensemble prediction for precipitation, was introduced as an input to the model. This was compared against predictions using the storage prediction method to evaluate accuracy. The results suggest that introducing MEPS into Elastic Net can provide accurate forecasts with safe results from a flood control perspective.