Comparison of group method of data handling based genetic programming and back propagation systems to predict scour depth around bridge piers

Abstract This study introduces a new application of GMDH in the prediction of scour depth around a vertical pier. Two models of the GMDH network were developed using genetic programming and a back propagation algorithm. Genetic programming was performed in each neuron of the GMDH instead of performing the quadratic polynomial. In the second model of the GMDH, the quadratic polynomial was used in each neuron of the network as a transfer function, and a back propagation algorithm was used for training of the network. Six effective parameters including pier diameter, flow velocity, flow depth, medium diameter of bed material, standard deviation of bed grain size and fluid dynamic viscosity were considered for prediction of the scour depth. Results of two GMDH networks were compared with results of several traditional equations. From result performances, it was found that although GMDH-GP is very time-consuming and more complicated, this proposed method has provided a better prediction of scour depth than GMDH-BP in training and testing stages. A sensitivity analysis was performed for the GMDH-GP model and the results indicated that the ratio of pier diameter to flow depth is the most significant parameter regarding scour depth. In particular, the GMDH network proved very effective in comparison with traditional equations.