A comparative evaluation of machine learning approaches for erosion prediction in dam-break problems

This study presents a novel integration of empirical erosion models optimized using machine learning methods in the finite volume solution of dam-break flows over erodible beds. Computational hydraulics and machine learning approaches are implemented and compared in modeling and simulation of erosio...

Celý popis

Uloženo v:

Podrobná bibliografie
Vydáno v:	European physical journal plus Ročník 140; číslo 10; s. 976
Hlavní autoři:	Al-Ghosoun, Alia, Gumus, Veysel, Seaid, Mohammed
Médium:	Journal Article
Jazyk:	angličtina
Vydáno:	Berlin/Heidelberg Springer Berlin Heidelberg 14.10.2025 Springer Nature B.V
Témata:	Accuracy Applied and Technical Physics Atomic Bed load Complex Systems Condensed Matter Physics Dam failure Datasets Entrainment Environmental impact Finite volume method Fluid flow Forecasting techniques Grain size Hydraulics Machine learning Mathematical and Computational Physics Mathematical models Molecular Neural networks Numerical analysis Optical and Plasma Physics Optimization Physics Physics and Astronomy Regular Article Researchers Sediment transport Shallow water Shallow water equations Soil erosion Support vector machines Theoretical Unmanned aerial vehicles Water
ISSN:	2190-5444, 2190-5444
On-line přístup:	Získat plný text
Tagy:	Přidat tag Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!

Popis
Shrnutí:	This study presents a novel integration of empirical erosion models optimized using machine learning methods in the finite volume solution of dam-break flows over erodible beds. Computational hydraulics and machine learning approaches are implemented and compared in modeling and simulation of erosional dam-break flows. Motivated by the need to enhance predictive accuracy, we employ seven empirical entrainment flux formulae and a well-balanced finite volume method to generate datasets with multiple input variables. The mathematical model consists of the nonlinear shallow water equations involving sediment transport and bedload terms. The computed results are used to train and test the considered machine learning models, including Bayesian neural networks (BNNs), K-nearest neighbors (KNN), M5 trees, multivariate adaptive regression splines (MARS), multiple linear regression (MLR), and support vector machines (SVMs). Quantitative evaluations using R 2 , Nash-Sutcliffe efficiency (NSE), and normalized root mean square error (nRMSE) revealed that SVM (with R 2 = 0.99, NSE = 0.99, nRMSE = 0.0245) and BNN (with R 2 = 0.98, NSE = 0.98, nRMSE = 0.035) significantly outperformed other models, with SVM slightly better during validation and testing. Concluding, this methodology optimizes existing empirical models with machine learning, improving prediction reliability for erosional dam-break flows. These findings are very important for hydraulics engineering by providing improved tools for accurate modeling and efficient simulation of sediment transport problems and thus have the potential to support practical applications in the field.
Bibliografie:	ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 14
ISSN:	2190-5444 2190-5444
DOI:	10.1140/epjp/s13360-025-06929-2