Studying the Accuracy and Generalizability of Different Estimation Methods of Shear Wave Velocity

In recent decades, due to the lack of shear velocity logs and the high cost of acquiring shear wave data in most wells, the use of different methods for estimating shear velocity is inevitable. In this research, different estimation methods of shear velocity, including Brocher, Nasrnia–Falahat model, Artificial neural-network, Mamdani inference system and Anfis were used. In the well X (training well), the correlation of the petrophysical logs with reference to measured shear velocity was calculated and it was distinguished that the compressional wave velocity, density and porosity parameters have a higher correlation than other parameters with the measured shear velocity. In this study, the mean squared error (MSE) and its root mean square error (RMSE) were used to evaluate and compare the different estimation methods in the well X. Since the data from this well was used to train the different methods, the accuracy and generalizability of these estimation methods were investigated in another well (test well). By studying and comparing the results of different methods of estimation in the test well, it was found that the method based on the Nasrnia–Falahat (rock physics model-based) demonstrated higher accuracy and generalizability compared to other shear velocity estimation methods. The results suggest that the mentioned intelligent methods, unlike the Nasrnia–Falahat model, experienced a significant decrease in generalizability under varying environmental conditions. In contrast, the Nasrnia–Falahat model maintained its accuracy in the test well. Therefore, in the similar geology, as well as in case of having access to the limited amount of data, it is suggested to use appropriate rock physics relationships that cover the principal parameters in estimating shear velocity.