Wang et al. (2025) Spatiotemporal correction of decision variables using XGBoost for multi-objective intelligent scheduling rule extraction model in reservoir-lake flood control systems

Identification

• Journal: Environmental Modelling & Software
• Year: 2025
• Date: 2025-11-20
• Authors: Huili Wang, Bin Xu, Xiaolin Qin, Xinrong Wang, Jianyun Zhang, Guoqing Wang, Fubao Yang, Ping‐an Zhong, Ran Mo, Xuesong Yang
• DOI: 10.1016/j.envsoft.2025.106795

Research Groups

• College of Hydrology and Water Resources, Hohai University, Nanjing, China
• National Key Laboratory of Water Disaster Prevention, Hohai University, Nanjing, China
• Key Laboratory of Hydrologic-Cycle and Hydrodynamic-System of Ministry of Water Resources, Hohai University, Nanjing, China
• State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering, Nanjing Hydraulic Research Institute, Nanjing, China
• Anhui Survey & Design Institute of Water Resources & Hydropower, Co., Ltd., Hefei, China

Short Summary

This study introduces a Spatiotemporal Correction using XGBoost (SC-XGB) technique to extract intelligent multi-objective scheduling rules for reservoir-lake flood control systems, addressing challenges in reducing spatiotemporal errors and improving Pareto frontier simulation quality. The SC-XGB model demonstrates enhanced accuracy and generalization in the Chaohu Basin, significantly improving outflow prediction, reducing water balance errors, and decreasing relative hypervolume error compared to the standard XGB model.

Objective

• To develop a spatiotemporal correction of decision variables technique using XGBoost (SC-XGB) for extracting intelligent multi-objective scheduling rules in reservoir-lake flood control systems, aiming to reduce spatiotemporal errors and improve the quality of Pareto frontier simulation.

Study Configuration

• Spatial Scale: Chaohu Basin, China (a specific river basin with a reservoir-lake system).
• Temporal Scale: Flood control scheduling, implying operational time scales (e.g., daily, hourly decisions) for real-time flood management.

Methodology and Data

• Models used: XGBoost (XGB), Spatiotemporal Correction using XGBoost (SC-XGB). A two-stage scheduling rule framework is designed, and a spatiotemporal correction loss function is introduced. Bayesian optimization with cross-validation is employed for hyperparameter tuning.
• Data sources: Implied from "reservoir-lake flood control systems" and "real-time changes in situations such as rainfall, floods, engineering, and disaster conditions" are hydrological observations (e.g., rainfall, inflow, water levels, outflow) and operational data for reservoirs.

Main Results

• The SC-XGB model improved the average Nash-Sutcliffe Efficiency (NSE) of outflow prediction by 1.89 % compared to the XGB model.
• It reduced the Water Balance Mean Error range of Chaohu Lake by 27.93 % compared to the XGB model.
• It decreased the Relative Hypervolume Error by 21.51 % compared to the XGB model.
• The SC-XGB model enhances both accuracy and generalization for intelligent scheduling in flood management systems.

Contributions

• Proposes a novel Spatiotemporal Correction of decision variables technique using XGBoost (SC-XGB) for extracting intelligent multi-objective scheduling rules.
• Designs a two-stage scheduling rule framework to reduce model complexity in multi-objective optimization.
• Introduces a spatiotemporal correction loss function to mitigate cumulative water balance constraint violation errors.
• Demonstrates significant improvements in outflow prediction accuracy, water balance error reduction, and Pareto frontier simulation quality compared to traditional XGB models in a complex reservoir-lake flood control system.

Funding

No funding information is provided in the excerpt.

Citation

@article{Wang2025Spatiotemporal,
  author = {Wang, Huili and Xu, Bin and Qin, Xiaolin and Wang, Xinrong and Zhang, Jianyun and Wang, Guoqing and Yang, Fubao and Zhong, Ping‐an and Mo, Ran and Yang, Xuesong},
  title = {Spatiotemporal correction of decision variables using XGBoost for multi-objective intelligent scheduling rule extraction model in reservoir-lake flood control systems},
  journal = {Environmental Modelling & Software},
  year = {2025},
  doi = {10.1016/j.envsoft.2025.106795},
  url = {https://doi.org/10.1016/j.envsoft.2025.106795}
}