Shear wave velocity prediction of shale oil formations based on machine learning and improved rock physics model

C onventional shear w ave（S ⁃ w ave）velocity prediction m ethods include em pirical form ulas and rock physics m odel m ethods. T he form er is suitable for reservoirs w ith relatively sim ple rock m ineral com positions，and it is affected by areas and som e other factors. T herefore，it is difficult to be w idely applied for different for⁃ m ations and has low prediction accuracy. T he latter requires selecting appropriate rock physics m odels based on different situations，so as to achieve the expected goals. M ost m achine learning m ethods for S ⁃ w ave velocity prediction aredriven by pure data，and the quality and quantity of the dataset directly determ ine the accuracy of the S ⁃ w ave velocity prediction m odel，w hich are in lack of sufficient physical insights. T herefore，based on the deep neural netw ork（D N N）m ethods，this paper assum es that the m athem atical form of w ave propagation equa ⁃ tions for the reservoir in the study area is know n，but the elastic param eters are unknow n and are learned through a D N N training on the basis of w ell logging data，so as to establish the w ave propagation equations of the target layer. T he corresponding com pressional w ave（P ⁃ w ave）and S ⁃ w ave velocities are obtained w ith the plane w ave analysis m ethod to connect the neural netw orks and the theoretical m odel. In addition，to address the shortcom ings of the conventional X u ⁃ W hite m odel，an im proved rock physicsm odel for S ⁃ w ave velocity pre⁃ diction is proposed by considering the pore aspect ratio varying with depth. By using the adequate well logging data in the study area，the established DNN model and the improved rock physics model for S ⁃ wave velocity prediction are used to predict the S⁃wave velocity，and the results are compared with the conventional Xu⁃White model. It shows that both the DNN model and the improved rock physics model can help obtain high⁃precision S⁃ wave velocity prediction results，and the former has better prediction performances. © 2024 Editorial office of Oil Geophysical Prospecting. All rights reserved.