Neuro-genetic Approach to Predict Scour Depth Around Vertical Bridge Abutment

Scour is caused by the erosive action of flowing water. Although, different researchers have proposed various empirical models to predict the equilibrium local scour depth around bridge abutment, these are suitable to a particular abutment condition. In this study, an integrated model that combines genetic algorithms (GA) and multilayer perceptron (MLP) network, a class of artificial neural network (ANN), is developed to estimate the scour depth around vertical bridge abutment. The equilibrium scour depth was modeled as a function of four affecting parameters of scour, abutment length, median grain size, approaching flow depth, and average approach flow velocity, and these parameters are considered as input parameter to the MLP model. The efficiency of the developed models is compared with the empirical equations over a dataset collected from literature. The MLP is found to outperform the empirical equations for the dataset considered in the present study. The performance of the best case MLP is further improved by applying GA for weight initialization. The results indicate that the GA-based MLP is more effective in terms of accuracy of predicted results and is a promising approach compared to MLP as well as the previous empirical approaches in predicting the scour depth at bridge abutments.