Wu et al. (2026) Physical-guided hybrid modelling framework for hydropower forecasting in non-stationary cascade reservoir systems

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-06-18
• Authors: Nan Wu, Liangang Chen, Zhongrui Ning, Yong Shi, Jerker Jarsjö, Ke Zhang
• DOI: 10.1016/j.jhydrol.2026.135894

Research Groups

• Nanjing Hydraulic Research Institute (NHRI)
• The National Key Laboratory of Water Disaster Prevention, Hohai University
• Department of Physical Geography, Stockholm University

Short Summary

This paper presents a physical-informed hybrid framework for hydropower forecasting in non-stationary cascade reservoir systems. The proposed LSTM-PHY model outperforms other methods by achieving high predictive skill and providing physically meaningful insights into cascade regulation.

Objective

• Investigate the feasibility of using physical-informed deep learning models for hydropower forecasting in complex cascade reservoir systems.

Study Configuration

• Spatial Scale: Cascade reservoir system (Yalong River)
• Temporal Scale: Monthly to annual time scales

Methodology and Data

• Models used:
  • LSTM (Long Short-Term Memory) neural network
  • BFAST (Bivariate Empirical Mode Decomposition and Trend Removal) for trend, seasonal, and residual component decomposition
• Data sources:
  • Hydropower generation data from Ertan and Jinping-I reservoirs
  • Climate and hydrological data from the Yalong River basin

Main Results

• The proposed LSTM-PHY model achieved high predictive skill (NSE = 0.90) for generation discharge forecasting in both calibration and validation periods.
• BFAST decomposition produced physically consistent components, capturing long-term evolution of regulation strategies, monsoon-driven inflow patterns, and high-frequency variability induced by short-term scheduling decisions.

Contributions

• The proposed LSTM-PHY model improves predictive skill while providing a physically meaningful multi-scale view of cascade regulation, supporting hydropower forecasting and adaptive management under climate change.
• This study demonstrates the potential of physical-informed deep learning models for complex hydropower systems, offering new insights into cascade reservoir operation and management.

Funding

• National Key Research and Development Program of China (Grant No. 2021YFA1402200)
• National Natural Science Foundation of China (Grant Nos. 52179219 and 52209221)

Citation

@article{Wu2026Physicalguided,
  author = {Wu, Nan and Chen, Liangang and Ning, Zhongrui and Shi, Yong and Jarsjö, Jerker and Zhang, Ke},
  title = {Physical-guided hybrid modelling framework for hydropower forecasting in non-stationary cascade reservoir systems},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.135894},
  url = {https://doi.org/10.1016/j.jhydrol.2026.135894}
}