Application and Assessment of SVM Algorithm to Simulate the Geometry of Scour Hole Downstream of a Siphon spillway

Accurate simulation of scour hole’s geometric specifications, created due to flow jet collision to plunge pool bed, has a significant role in the design, construction and performance of weirs and energy dissipater structures. In this research, a comparative evaluation was done between experimental and numerical study of scour phenomena downstream of a siphon spillway using SVM algorithm with three sediment particles and three buckets angles for different magnitudes of flow discharge and tail water depths in a rectangular flume made from plexiglas. Three linear-independent dimensionless parameters were formed. Classic linear and non-linear regression model and FFBP and CFBP neural networks were applied more over the SVM algorithm to simulate the scour hole dimensions and were compared using five statistical indices as R, RMSE, MAE, N.S and DDR. A comparison of the quintuple accuracy indices showed that SVM algorithm had better performance than to the other models, especially classic regression ones. Among all parameters, SVM algorithm had acceptable prediction with higher and lower level of accuracy for "L" _"S" /"h" _"t" and "L" _"6" /"h" _"t" by quintuple above mentioned indices as (0.9954, 0.2056, 0.987,0.9851,7.855) and (0.8924,2.860,2.23,0.7587,1.405), respectively. Finally, a comparison was done among SVM algorithm and empirical predictors of scour hole depth. Comparison of accuracy indices showed that the SVM had a tangible and absolute advantage over empirical equations presented by other researchers.