Author(s): Mohammad Rezaie-Balf; Mohammad Najafzadeh; Ali Tafarojnoruz

Linked Author(s):

Keywords: Riprap; Stone-sizing relationships; Gene-expression programming; Evolutionary polynomial regression; Model tree; Overtopping flow

Abstract: Rock riprap is commonly used to protect levees, embankment dam, steep channels, and other structures being vulnerable to deteriorative erosion caused by overtopping flow. A review of the literature in this context indicates that over 24 riprap design expressions exist to predict the stone size in the riprap layer. However, each equation was originally derived on the basis of limited data sets under certain experimental conditions. In this investigation, Gene-Expression Programming (GEP), Model Tree (MT), and Evolutionary Polynomial Regression (EPR) were evaluated to estimate riprap stone sizing by virtue of 102 experimental data sets. Efficiency and performance of GEP, MT, and EPR approaches for training and testing stages were analysed and discussed. Results analysis revealed that the EPR technique provided an accurate prediction of riprap sizing in testing stage compared with other selected data-driven models as well as the empirical equations. The robustness of the developed data-driven predictive techniques was verified through the external validation: the selected data-driven models are definitely valid, have the strong capability to predict the riprap stone size which was not achieved by a chance. The prediction uncertainties of the data-driven models were quantified and compared with the selected empirical equations.

DOI: https://doi.org/10.1080/15715124.2018.1437738

Year: 2018