IAHR Document Library

« Back to Library Homepage « Proceedings of the 39th IAHR World Congress (Granada, 2022)

A Study on Improving the Accuracy of Dam Inflow Prediction Based on Deep Learning

Author(s): Jae-Yeong Heo; Ye-Jin Lim; Deg-Hyo Bae

Linked Author(s): Deg-Hyo Bae

Keywords: Dam Inflow Prediction; Deep Learning; Lead Time; Rainfall-Runoff; Multi-step-ahead

Abstract: Accurate and reliable prediction of dam inflow is essential in terms of the mitigation of water disaster risk and water resources management. Nevertheless, the dam inflow prediction based on the physical-based models requires a large amount of computation time. This is a one of the reasons that cannot meet the needs of real-time dam inflow predictions. Also, there are difficulties in capturing features of the nonlinear and complex processes of rainfall-runoff. For this reason, this study proposed a data-driven method to improve the accuracy of real-time inflow prediction. This proposed method has a capability for multi-step-ahead inflow predictions considering the features of forecast period. The study area is the Soyang River Dam basins in Han river located in South Korea that has a large number of hourly data. In order to evaluate the proposed method, quantitative measures are used for 1 to 6 hour-ahead predictions. The results showed that the proposed method improve the accuracy of dam inflow predictions when the forecast time step is increased. Also, when considering the forecast rainfall, the performance outperformed compared to the cases that only consider past period for input sequences. The hydrographs for 6 hour-ahead reflected well the features of shape and peak flow for flood event. We concluded that the proposed method that consider the forecast data can improve the reliability of dam inflow prediction.


Year: 2022

Copyright © 2024 International Association for Hydro-Environment Engineering and Research. All rights reserved. | Terms and Conditions