Evaluation of regular wave scour around a circular pile using data mining approaches

Abstract

An accurate estimation of scour depth around a pile is very difficult due to the complex behavior of flow around a pile structure on an erodible bed. In the current study, Regression Trees (RT) and Artificial Neural Networks (ANNs) as remedy data mining approaches are suggested to estimate the scour depth due to regular waves. These approaches were used to predict normalized scour depth as a function of two separate sets of parameters: (i) dimensional parameters and (ii) dimensionless parameters. The ANN trained by dimensional parameters provides more accurate results compared to that trained by dimensionless parameters. As opposed to the ANN model, the RT model based on dimensionless input parameters predicts normalized scour depth outperformed the one based on dimensional inputs. In addition, these models outperformed the existing empirical formulae. A committee model based on the geometric mean of the results of RT and ANN (developed by dimensionless parameters) is presented as the best model. To determine the relative importance of input parameters in the prediction of the scour depth, a sensitivity analysis was then performed and it was found that the Keulegan–Carpenter number ( KC) was found to be the most important one. The error statistics for two classes of K C ( K C < 10 and K C > 10 ) indicated that the suggested approach performs better in the range of K C < 10 for the prediction of dimensionless scour depth.