Neural network based prediction of shear wave velocity for soils

Abstract

The paper presents a study on the prediction of the shear wave velocity (Vs) for all soils using ANN model from cone penetration test data. The input parameters in the obtained of ANN models were vertical effective stress, sleeve friction, cone tip resistance and output was the shear wave velocity. These parameters were considered to construct the ANN model which has 3-2-1 topology in prediction of shear wave velocity. Further, utilizing the RMSE, MAE, MSE, MAPE, determination coefficient (R2) and correlation coefficient (r) for training and testing data set, performance of artificial neural networks was investigate utilizing different activation function. The created ANN model had an adequate precision. The contribution of input parameters on prediction of shear wave velocity was determined by sensitivity analysis and it is found that the contribution of vertical effective stress was 84% on the output shear wave velocity. It was much more as compared to other parameters. Therefore, the ANN model having topology of 3-2-1 was better as compared to the one developed using multi linear regression analysis (MLRA) as well as previously reported correlations in literature to estimate the shear wave velocity. Finally, an equation was proposed for the ANN model based on trained of weights and biases obtained.