Data-intelligence approaches for comprehensive assessment of discharge coefficient prediction in cylindrical weirs: Insights from extensive experimental data sets

•Kernel-based models such as SVM, GPR, KELM, and KRR were employed, alongside neural network-based models including FFBP, CFBP, GRNN, and RBFN for discharge coefficient prediction of cylindrical weir.•855 experimental data points from different types of cylindrical weir used to feed the utilized mod...

Full description

Saved in:
Bibliographic Details
Published in:Measurement : journal of the International Measurement Confederation Vol. 233; p. 114673
Main Authors: Roushangar, Kiyoumars, Shahnazi, Saman, Mehrizad, Amir
Format: Journal Article
Language:English
Published: Elsevier Ltd 30.06.2024
Subjects:
Artificial neural networks
Civil Engineering
Cylindrical weir
Discharge coefficient
Discharge prediction
Engineering and Technology
Geotechnical Engineering and Engineering Geology
Geoteknik och teknisk geologi
Kernel-based models
Samhällsbyggnadsteknik
Teknik
Cylindrical weir
Discharge prediction
Kernel-based models
Discharge coefficient
Artificial neural networks
ISSN:0263-2241, 1873-412X
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•Kernel-based models such as SVM, GPR, KELM, and KRR were employed, alongside neural network-based models including FFBP, CFBP, GRNN, and RBFN for discharge coefficient prediction of cylindrical weir.•855 experimental data points from different types of cylindrical weir used to feed the utilized models.•The predictive capability of discharge coefficient for various types of cylindrical weirs was discussed.•The prediction performance of employed data-intelligence techniques was assessed under varied hydraulic conditions. The present study collected a wide range of experimental data samples, including 855 records from various types of cylindrical weirs under diverse hydraulic conditions, setting the stage for an in-depth analysis for modeling of discharge coefficient. For this purpose, the prediction capability of four kernel-based methods (SVM, GPR, KELM, and KRR) and four neural network-based methods (FFBP, CFBP, GRNN, and RBFN) was thoroughly investigated. Results reveal that among the neural network methods, FFBP demonstrates superior performance with a correlation coefficient (R) of 0.973, Nash-Sutcliffe Efficiency (NSE) of 0.942, and Mean Squared Error (RMSE) of 0.014, particularly excelling at high flow rates. Conversely, KRR emerges as the top-performing kernel-based method with R = 0.901, NSE = 0.811, and RMSE = 0.058. Moreover, the prediction process unveils challenges associated with hydraulic conditions, notably the presence of upstream ramps, which substantially diminish the accuracy of modeling results by 112 %.
ISSN:0263-2241
1873-412X
DOI:10.1016/j.measurement.2024.114673