Prediction model of local scour depth of bridge piers based on LS-SVM

Abstract

Local scour of pier foundation soil is one of the main causes of bridge failure. The scour mechanism around the pier foundation is complex. The current code mainly uses empirical formulas to predict the scour depth of the pier, and the prediction results are generally too discrete. In order to accurately predict the local scour depth of bridge piers, 337 sets of model scour test data were collected in this paper. The standardized method was used to process the data dimensionlessly, and the Pearson correlation analysis method was used to analyze the correlation of the experimental data. It is concluded that the pier diameter, water flow depth, water flow velocity, median particle size and particle size standard deviation are the main influencing factors of local scour depth of bridge piers. The sensitivity analysis method is used to analyze the sensitivity of the five parameters and analyze their influence on the local scour depth of the pier. A prediction model of local scour depth of bridge piers based on least squares support vector machine (LS-SVM) is proposed. The results show that the prediction results of the model are obviously better than the calculation results of the current specification. After the dimensionless treatment, the coefficient of determination of the prediction model is increased from 0.624 to 0.824. The predicted value of the local scour depth of the pier is in good agreement with the measured value, which can provide reference for bridge design and safe operation.