Generalized Neural Networks and Multi Layer Perceptron for Prediction of Equilibrium Scour Depth Around Bridge Piers

Perceptron and generalized regression neural network (GRNN) have been used to predict the maximum depth of scour around bridge piers. Levenberg-Marquardt and Momentum as training algorithms and Sigmoid and Tanh as activation functions have been used in this study. Many studies have been conducted recently using the artificial intelligence techniques to predict bridge pier scour. Application of the circular pier shape and experimental data in place of the real data are the weaknesses of most of the previous stidies. Therefore، the rectangular-based piers with rounded edges، and the sharp-edged piers have been studied in the present research in addition to the cylindcal piers. Furthermore، 475 real data have been used to build and verify the neural models. In contrast to the studies in which the trial and error were used to determine the number of hidden layers، the genetic algorithm was used in this study to determine hidden layer neurons. The result of all neural models indicated that the GRNN model estimates equilibrium scour depth more accurately than the other models. To examine the accuracy of the GRNN model، some popular empirical models such as the Laursen and Toch، Shen، Breusers et al.، Froehlich، Melville and Melville، and Chiew have also been used. Although the Froehlich model provided a better accuracy than those of the others، the result developed by the GRNN model were in better agreement with the real data. The sensitivity analysis indicated that the mean velocity had a greater influence on equilibrium scour depth than the other independent parameters.