Reservoir Inflow Forecasting of The Bhumibol Dam Using XGBOOST Algorithm
Issued Date
2023-01-01
Resource Type
ISSN
25217119
eISSN
2521716X
Scopus ID
2-s2.0-85187720538
Journal Title
Proceedings of the IAHR World Congress
Start Page
1711
End Page
1719
Rights Holder(s)
SCOPUS
Bibliographic Citation
Proceedings of the IAHR World Congress (2023) , 1711-1719
Suggested Citation
Dornpunya P., Musor H., Rittima A., Kraisangka J. Reservoir Inflow Forecasting of The Bhumibol Dam Using XGBOOST Algorithm. Proceedings of the IAHR World Congress (2023) , 1711-1719. 1719. doi:10.3850/978-90-833476-1-5_iahr40wc-p1611-cd Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/97712
Title
Reservoir Inflow Forecasting of The Bhumibol Dam Using XGBOOST Algorithm
Author(s)
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
Recent research has been devoted to enhancing the predictive performance of hydrological forecasting models through Machine Learning (ML) techniques, aiming for successful local government decision-making in water resources management and planning in times of crisis and normal situations. This study focuses on applying ML to forecast reservoir inflow at Bhumibol (BB) dam, the primary water source in the Ping River Basin, northern Thailand, with two operation centers, Kamphaeng Phet and Lamphun Water Resources Management Operation Centers, supervise and manage water in the Ping River Basin area. The eXtreme Gradient Boosting (XGBoost) algorithm, an ensemble ML algorithm based on decision trees, was utilized for forecasting daily reservoir inflow using the R programming language. The model's training and testing phases employed inflow and rainfall data spanning from 2000 to 2022 as key forecasting inputs. The XGBoost model was trained and tested while adjusting various parameters, including the ratio of training-to-testing datasets, learning rates, average inflow, rainfall at delayed time steps (1, 3, and 7 days or t-1, t-3, and t-7), maximum iteration number, and early stopping rounds. Statistical performance such as coefficient of determination (R-square) and Root Mean Square Error (RMSE) were used to evaluate the forecasting models' effectiveness. Validation results indicate that the XGBoost algorithm can replicate the reservoir inflow pattern and yield robust forecasting results, achieving a high R-square value of 0.8898 and a low RMSE of 7.2964. However, a notable underestimation of peak inflows was observed, leading to a volume error of –25.58 MCM. Therefore, optimizing the ML parameters remains crucial to accurately capture extreme reservoir inflow values, which are pivotal for effective water resource management in anticipation of hydrological events. In particular, precise forecasting data will be utilized to strengthen the capability of the Kamphaeng Phet and Lamphun Water Resources Management Operation Centers in these challenging climate times.