Chen et al. (2026) A LSTM reservoir outflow model enhanced by operation chart derives the representative reservoir operation schemes

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-09-08
• Authors: Runting Chen, Dagang Wang, Yiwen Mei, Yongen Lin, Jinxin Zhu, Xiaoxing Qi
• DOI: 10.1016/j.ejrh.2026.103923

Research Groups

• School of Geography and Planning, Sun Yat-sen University, Guangzhou, China
• Carbon-Water Research Station in Karst Regions of Northern Guangdong, School of Geography and Planning, Sun Yat-sen University, Guangzhou, Guangdong, China
• Guangdong Key Laboratory for Urbanization and Geo-simulation, Sun Yat-sen University, Guangzhou, China
• School of Government, Sun Yat-sen University, Guangzhou, China

Short Summary

This study develops a knowledge-guided Long Short-Term Memory (KG-LSTM) model to simulate reservoir outflow by incorporating the reservoir operation chart (ROC). The KG-LSTM achieves superior performance compared with the standard LSTM, indicating that incorporating empirical reservoir operation knowledge improves the reservoir outflow simulation.

Objective

• To represent the impact of reservoir operation on streamflow dynamics
• To extract empirical ROC knowledge to enhance LSTM outflow simulation
• To extract unique reservoir operation insights from KG-LSTM

Study Configuration

• Spatial Scale: The study focuses on the Xinfengjiang Reservoir in the Dongjiang River Basin, China.
• Temporal Scale: The study uses monthly-scale data from January 1961 to December 2006.

Methodology and Data

• Models used: Long Short-Term Memory (LSTM) network, Classification and Regression Trees (CART), Shapley Additive exPlanations (SHAP)
• Data sources: Reservoir operation data, streamflow data from Heyuan hydrological station

Main Results

• The KG-LSTM achieves superior performance compared with the standard LSTM, indicating that incorporating empirical reservoir operation knowledge improves the reservoir outflow simulation.
• The interpretability analysis identifies previous-month outflow as the most influential factor in current outflow determination.
• Six representative and situation-dependent operation schemes are identified, corresponding to three major reservoir functions: impoundment, power generation, and water replenishment.

Contributions

• This study provides a framework that enhances both accuracy and interpretability of data-driven reservoir modeling, offering a new avenue for operation knowledge extraction and validation.
• The study demonstrates the effectiveness of incorporating empirical ROC knowledge into LSTM models to improve outflow simulation performance.

Funding

• This research was supported by the National Natural Science Foundation of China (Grant No. 51879224)
• Guangdong Provincial Key Laboratory of Urbanization and Geo-simulation (Grant No. 2021A0505030003)
• The Fundamental Research Funds for the Central Universities (Grant No. 21ykpy37)

Citation

@article{Chen2026LSTM,
  author = {Chen, Runting and Wang, Dagang and Mei, Yiwen and Lin, Yongen and Zhu, Jinxin and Qi, Xiaoxing},
  title = {A LSTM reservoir outflow model enhanced by operation chart derives the representative reservoir operation schemes},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2026.103923},
  url = {https://doi.org/10.1016/j.ejrh.2026.103923}
}