Enhancing breakout identification in geomechanical modeling: using fullset logs with machine learning in carbonate reservoirs

In the present research, a comparison between random forest and stacking ensemble learning approaches is presented to identify breakout zones in carbonate formations based on machine learning techniques. Breakout zones play a very significant role in hydraulic fracturing and wellbore stability within the frame of geomechanical modeling. This work evaluated the efficiency of machine learning approaches in breakout zone prediction for four wells in two different fields using the petrophysical logs such as gamma ray (GR), sonic transit time (DT), density log (RHOB), formation evaluation photoelectric factor (PEF), deep resistivity log (RT), neutron porosity (NPHI), caliper, bit size and formation microresistivity image (FMI) logs. Results showed that the accuracy ranking of all test wells using the Stacking Ensemble method ranged from 0.86 to 0.89, while those using the Random Forest ranged between 0.55 and 0.84. In general, results have indicated that on scope and accuracy, the Stacking Ensemble method outperformed the Random Forest against a well-defined circumstance range. Geomechanical modeling has illustrated that intelligent approaches for breakout prediction enhance the accuracy of the geomechanical models. This work will finally illustrate how machine learning can enhance breakout zone detection, further enhance geomechanical modeling and optimize oil and gas development by ensuring stability in the wellbore.