Modeling Siltation of River Channels Using the Physics-Informed Neural Networks Method and Numerical Simulation
Abstract
In this study, we developed a Physics-Informed Neural Network (PINN) model integrated with the Navier-Stokes equations to simulate sediment transport and siltation in river channels. The model was designed to predict sediment deposition in complex hydrodynamic environments, using real-world data from the Syrdarya River and Shardara Reservoir and remote sensing data from Sentinel-1 and Sentinel-2 satellites. The PINN demonstrated a 10-15% reduction in computation time compared to traditional numerical methods like Ansys Fluent while maintaining high accuracy in predicting flow velocity, pressure, and sediment accumulation. By integrating physical laws and real-time data, this model has the potential to significantly improve flood management and river channel design, offering a faster, more adaptive approach to environmental hydrodynamic modeling.