Application of Machine Learning Model for Estimating the Geomechanical Rock Properties Using Conventional Well Logging Data

Building a geomechanical model for reservoir rocks is crucial for oil and gas operations. It is essential to solving multiple designs like wellbore stability for drilling operations, hydraulic fracturing, and sand production prediction for production operations. The best method to build a geomechanical model is to measure in the lab or calculate it from the dipole sonic log. However, it cannot be practically done routinely due to the high cost of logging and processing the dipole sonic logs. With the training of a machine learning model using conventional logging data and dipole sonic logs and static geomechanical measurements in the lab, a machine learning tool is provided to predict the dipole sonic logs and build a geomechanical model using routinely recorded well logs like gamma-ray, resistivity, neutron porosity, and density. The calculated minimum horizontal stress was calibrated practically with the derived closure pressure derived from several diagnostic fracture injection tests. This paper provides a practical implementation of a theory-controlled data learning model. It introduces an innovative way to build a calibrated machine learning tool that can predict shear and compressional wave velocities and estimate the rock mechanical properties using the regular conventional well logging data, which are helpful for oil and gas operations.