Prediction of maximum scour depth downstream of bed sills using Support Vector Machines

Hydraulic grade-control structures (e.g. drops, bed sills, check dams, etc) havebeen widely used in rivers with low stability and high erosion especially inmountain streams. For proper and safe design, prediction of maximum scour depthdownstream of grade-control structures in rivers has vital importance. For thisreason, researchers have focused on developing simple and accurate empiricalequations in form of non-linear regression models for predicting scour depth.However, accuracy of regression models is limited in general and they have goodapplicability just for their experimental data. In this paper, Support VectorMachines (SVM), in forecasting the scour depth downstream bed sills is applied.226 experimental data sets from literatures with different hydraulic and sedimentconditions and at clear-water scouring have been used. Comparison of results intesting phase showed that outcome from the support vector machines withcoefficient of determination of 0.96, root mean square errors of 0.539 and meanabsolute errors of 0.4 suggest a better performance to existing regression comparedequations. Also, it is found that from 5 effective input dimensionless parametersincluding a/Hs, D50/Hs,, S00 50 a / D and s L / H, only the first three parametershad greater effects on modeling maximum scour depth at bed sills and theremaining may be omitted.