Huang et al. (2026) Configurable physics-informed operator network for real-time multi-scenario hydrodynamics in river networks

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-01-28
• Authors: Bingxi Huang, Huiming Zhang, Sheng Jiang, Hongwu Tang, Saiyu Yuan, Xiao Luo, Zhaohui Chen
• DOI: 10.1016/j.jhydrol.2026.135042

Research Groups

• State Key Laboratory of Water Disaster Prevention, Hohai University, Nanjing 210024, China
• State Key Laboratory of Water Cycle and Water Security, Hohai University, Nanjing 210024, China
• College of Water Conservancy and Hydropower Engineering, Hohai University, Nanjing 210024, China
• Key Laboratory of Hydrologic-Cycle and Hydrodynamic-System of Ministry of Water Resources, Hohai University, Nanjing 210024, China
• School of Civil Engineering and Transportation, South China University of Technology, Guangzhou 510641, China
• School of Civil Engineering, The University of Sydney, NSW 2006, Australia

Short Summary

This paper introduces a physics-informed river operator network (PI-RONet) for real-time, multi-scenario hydrodynamic prediction in plain river networks, achieving high accuracy and significantly reducing computation time compared to traditional numerical models.

Objective

• To develop a configurable physics-informed operator network (PI-RONet) that provides real-time, accurate, and physically consistent hydrodynamic predictions for multi-scenario conditions in plain river networks, overcoming the computational expense of traditional models and the generalization/physical consistency limitations of purely data-driven approaches.

Study Configuration

• Spatial Scale: Plain river networks, large-scale.
• Temporal Scale: Real-time, unsteady, dynamic, multi-scenario.

Methodology and Data

• Models used: Physics-informed river operator network (PI-RONet), which integrates a configurable operator network (DeepONet/MIONet) as an encoder and a Physics-Informed Neural Network (PINN) as a decoder. The PINN embeds the Saint-Venant equations and junction equations into its loss function.
• Data sources: Large-scale, high-fidelity datasets generated by an automated Python-RAS workflow across multiple flow conditions.

Main Results

• PI-RONet achieved high accuracy (R² ≥ 0.8) for discharge in 99.5% of conditions and for water level in 90.9% of conditions under 1000 unsteady test boundary conditions.
• The model reduced multi-scenario simulation time from hours to seconds, demonstrating a nearly 5000-fold acceleration compared to traditional numerical models.
• Ablation experiments showed that MIONet improved water level prediction accuracy by reducing the Root Mean Square Error (RMSE) by 19.5% and shortened training time by 10.7%.
• The trained model was deployed as an interactive web platform, RivONet, showcasing its potential for real-time decision support and intelligent water management.

Contributions

• Proposes PI-RONet, a novel framework that combines configurable operator networks with physics-informed neural networks to enable real-time, multi-scenario, and physically consistent hydrodynamic prediction in river networks.
• Addresses critical limitations of existing methods, including the computational cost of numerical models and the lack of generalization and physical consistency in purely data-driven approaches.
• Develops an automated Python-RAS workflow for efficient generation of large-scale, high-fidelity datasets required for model training.
• Demonstrates significant computational acceleration (approximately 5000 times faster) while maintaining high accuracy compared to traditional numerical models.
• Provides insights into the benefits of MIONet over DeepONet for improved prediction accuracy and training efficiency in this context.
• Showcases the practical applicability of the model through its deployment as an interactive web platform (RivONet) for decision support.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Huang2026Configurable,
  author = {Huang, Bingxi and Zhang, Huiming and Jiang, Sheng and Tang, Hongwu and Yuan, Saiyu and Luo, Xiao and Chen, Zhaohui},
  title = {Configurable physics-informed operator network for real-time multi-scenario hydrodynamics in river networks},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.135042},
  url = {https://doi.org/10.1016/j.jhydrol.2026.135042}
}