Comparison of Three Soft Computing Methods in Estimating Apparent Shear Stress in Compound Channels

Apparent shear stress acting on a vertical interface between the main channel and floodplain in acompound channel serves to quantify the momentum transfer between sub sections of this crosssection. In this study, three soft computing methods are used to simulate apparent shear stress inprismatic compound channels. The Genetic Algorithm Artificial neural network (GAA), GeneticProgramming (GP) and Modified Structure-Multi Layer Perceptron (MS-MLP) are applied to about100 different data to predict apparent shear stress. The modelling procedure with three models wereextended and the best of each model was selected after each step. In modeling with the GAA and GPdifferent input combinations, fitness functions, transfer functions and mathematical functions wereinvestigated for obtaining the optimum combination. The results showed B/b, H/B, nf/nc and h/b asinput combination, fitness function MSE and transfer function tan-pur is the best combination for GAAmodel. The best GP model introduced with B/b, (H-h)/h, nf/nc and h/b as input variables, fitnessfunction MAE and ,,,,sin,cos,abs, sqrt, power as the mathematical function set. Finally, themost appropriate GAA, GP and MS-MLP models were compared to select the best of them inestimating apparent shear stress in compound channels. According to the results, MS-MLP improvedwith RMSE of 0.3654 over GAA with RMSE of 0.5326 and the GP method with RMSE of 0.6615.