Predicting shear wave velocity from conventional well logs with deep and hybrid machine learning algorithms

Abstract Shear wave velocity ( V S ) data from sedimentary rock sequences is a prerequisite for implementing most mathematical models of petroleum engineering geomechanics. Extracting such by analyzing finite reservoir cores very costly and limited. The high cost sonic dipole advanced wellbore logging service its implementation in few wells field has placed many limitations on geomechanical modeling. On the other hand, shear tends to be nonlinearly related influencing variables, making empirical correlations unreliable prediction. Hybrid machine learning (HML) algorithms are well suited improving predictions variables. Recent advances deep (DL) suggest that they too should useful predicting large gas oil datasets but this yet verified. In study, 6622 records two giant Iranian Marun (MN#163 MN#225) used train HML DL algorithms. 2072 independent another (MN#179) verify prediction performance based eight well-log-derived Input variables standard full-set recorded parameters conventional available older wells. predicts supervised validation subset with root mean squared error (RMSE) 0.055 km/s coefficient determination (R 2 0.9729. It achieves similar accuracy when applied an unseen dataset. By comparing results, it apparent convolutional neural network model slightly outperforms tested. Both HLM substantially outperform five commonly relationships calculating p Field Concerns regarding model's integrity reproducibility were also addressed evaluating field. findings study can lead development knowledge production patterns sustainability reservoirs prevention enormous damage geomechanics through better understanding instability casing collapse problems. Graphical abstract