Application of SVM Algorithm in Predicting Vertical Pier Scour Depth

Local scour around the foundation of marine and hydraulic structures is one of the most important factors in the instability and destruction of these structures. False prediction of scour depth around bridges has caused financial losses in plasticization and endangered many people's lives. Therefore, an accurate estimation of this complex phenomenon around the bridges is necessary. Also, since the formulas presented by different researchers relate to laboratory conditions, they are less true and less accurate in other situations. Recently, many researchers have tried to introduce new methods and models called soft calculations in predicting this phenomenon. In this research, 146 different laboratory data series (three different laboratory conditions) were analyzed using a backup vector machine to predict scour depth around the bridge head. These data are presented in the form of various combinations of input parameters  which, respectively, represent thickness under the slippery layer, Reynolds number, critical velocity, Shields parameter, velocity Shear, average speed, flow depth, the average diameter of the particles and diameter of the bridge. The parameters in two different scenarios (the mode with dimension and mode) were introduced into the SVM network and the results of this machine were compared with those obtained from the experimental formulas and relations presented in this study. The results showed that in the first scenario, the combination of No. 5 with input parameters () and in the second scenario, the combination No. 5 with input parameters  () for the test stage were selected as the best model. It was also concluded from the results that the scenario two (the state with dimension) in predicting the scour depth around the vertical single-pillar provided a more accurate estimate than the first scenario (barrier state). At the end, the sensitivity analysis was carried out on the parameters and the parameters D, U*, V were selected, respectively, as the most effective parameters