Skip to main content
2024 EWRI Congress Main banner
Back

2024 EWRI Congress

Hydraulics & Waterways

Development and Application of Machine Learning and Artificial Intelligence (ML/AI) Techniques in Computational Hydraulics - II

423 - Developing Surrogate Models for Efficient Levee Breach Flow Estimation

Wednesday, May 22, 2024
2:00 PM – 2:15 PM CT
Room: S 101 D
This study represents a significant advancement in the application of Machine Learning (ML) models as surrogates of hydraulic modeling, with a particular focus on estimating levee breach discharge and overtopping depth in rectangular channels. Multi-Layer Perceptron (MLP) and Gaussian Process Regression (GPR) models were trained by integrating results from a 2D numerical model of levee breaches into the ML loss function. The accuracy of the 2D numerical model was validated against experimental measurements. The input variables for the surrogate models include inlet discharge, downstream depth, breach width, and levee height, all non-dimensionalized to make the model scale-independent, enabling its applicability across various scenarios. The MLP achieved R-squared values of 0.99 and 0.97 for breach discharge and overtopping depth, respectively. Furthermore, the MLP model exhibited predictive capabilities with an R-squared value of 0.95 when extrapolating beyond the training data. Future research will extend this study to incorporate the impact of channel width and other cross-sectional parameters on breach outflow and overtopping depths. This expansion holds the promise of further enriching our understanding and predictive potential in the domain of physics-informed machine learning for hydraulic modeling, particularly in scenarios involving levee breaches.