Estimation of shear wave velocity using Gaussian process regression, multivariate regression and multilayer perceptron artificial neural network

Estimation of wave velocities is very important for designing geotechnical structures and modeling deep drillings. The purpose of this study is to estimate shear wave velocity (Vs) using Gaussian process regression (GPR), multilayer perceptron artificial neural network (MLP-ANN) and multivariate linear regression (MVLR) methods. In order to carry out this study, 14 rock blocks were prepared from the northwest of Damavand city and after being transferred to the laboratory, cores were extracted from them. In order to develop a predictive model, point load index, compressional wave velocity (Vp), porosity and density tests were performed on 61 rock core samples. Point load index, Vp, porosity and density were used as input parameters of models to predict Vs. The results of lithological studies showed that the studied sandstones are feldspathic litharnite and litharnite. The results showed that the ratio of Vp to Vs is equal to 1.70 on average. The results of the MLP-ANN showed that the highest accuracy of the models was obtained by using the Levenberg-Marquardt training algorithm. The most accurate models were obtained using this algorithm to estimate the Vs in neuron number 2 (optimal neuron). The GPR, MLP-ANN and MVLR predicted Vs with correlation coefficients of 0.97, 0.96 and 0.95, respectively. GPR method showed better performance in predicting Vs than other methods.