Rapid hydraulic prediction for side-channel spillways with a physics-informed data-driven framework

ABSTRACT

The design of side-channel spillways, while advantageous for reservoirs in complex terrain, especially those with high slopes, is hindered by the challenge of predicting their intricate three-dimensional flow. Traditional methods are too costly and inaccurate for modern rapid-design requirements. To address these limitations, this study uses high-fidelity numerical simulations to systematically analyze the coupled effects of discharge, side-slope coefficient, and bottom-width variation on key hydraulic parameters. This analysis informs the development of a physics-informed neural network (PINN) model that integrates the Navier-Stokes equations as physical constraints, enabling the rapid and accurate prediction of critical flow characteristics such as velocity, pressure, and Froude number. Results show that increasing the side-slope coefficient and the rate of bottom-width variation suppresses transverse circulation, reduces turbulent kinetic energy, and promotes a more uniform velocity distribution. The PINN model achieves high prediction accuracy with computational efficiency far surpassing conventional computational fluid dynamics (CFD). Moreover, the model's error patterns align with those of the numerical simulations, validating its robustness. This research establishes a novel, efficient framework for optimizing side-channel spillway design, with significant potential for extension to other hydraulic engineering applications to enhance design efficiency and safety.