Predicting dynamic shear wave slowness from well logs using machine learning methods in the Mishrif Reservoir, Iraq

Shear wave slowness is needed in reservoir characterization for seismic modeling, amplitude variation analysis and determination of rock elastic properties. Conventional methods such as multi-array polarization acoustic measurements as well as Dipole Shear Sonic Imager DSI are employed for measuring the shear slowness directly. However, data from these tools aren't available in all wells, particularly older ones, and in some cases, due to high cost and technical difficulties (for instance, in cased hole). Consequently, statistical techniques and machine learning applications are quick and can be applied with the appropriate algorithm to obtain this acoustic parameter. This provides reasonable results, saves time and cost of running experiments and well loggings. This research, therefore, provides an improved methodology via Geolog software's Facimage module to predict the dynamic shear slowness for the Mishrif formation, Iraq. Five different techniques were employed: self-organizing map (SOM), ascendant hierarchical clustering (AHC), dynamic clustering (DYN), artificial neural networks (ANNs), and multi-resolution graph-based clustering (MRGC). The predictions yielded root mean square error (RMSE) of 5.778, 6.914, 6.011, 4.845 and 2.897, while the R - squared (R-2) values are 0.838, 0.766, 0.824, 0.888, and 0.9591 for SOM, AHC, DYN, ANN and MRGC methods, respectively. Implying that the MRGC yielded a predicted shear wave slowness with best match. The application of this advanced statistical approach in reservoir characterization will help to exploit the huge well log data available, consequently, saving time for reservoir description and production cost as well as improved ultimate recovery.

Automated summary

This research aims to predict the dynamic shear slowness of the Mishrif formation in Iraq using the Geolog software's Facimage module and five different statistical techniques. The results show that the multi-resolution graph-based clustering (MRGC) method provides the best match with the actual data. The application of this advanced statistical approach helps to utilize well log data, save time and cost, and improve the ultimate recovery.