Machine Learning Assisted State-of-the-Art-of Petrographic Classification From Geophysical Logs

In the E&P industry, accurate lithology classification is an essential task for successful exploration and production. Geophysical logs provide high-resolution petrophysical properties, but core logging is expensive and traditional techniques may not accurately classify lithologies. We demonstrated a comparative analysis of six ML algorithms: k-nearest neighbor (kNN), support vector machine (SVM), decision tree (DT), random forest (RF), extreme gradient boosting (XGBoost) and artificial neural network (ANN) for the prediction of lithologies from geophysical logs. Here we analysed the wireline logs of eight wells associated with the petroliferous Lakadong-Therria formation of the Bhogpara oil field of the Assam-Arakan Basin. This formation contains eight typical lithologies, namely clay stone, sand stone, calcareous sandstone, shale, calcareous shale, carbonaceous shale, coal and limestone. Performance of the ML algorithms were evaluated through accuracy, precision, recall, F1-score and receiver operating characteristic (ROC) curve. During the training and test phases, the computed overall accuracy of the predicted ML modes exceeded 82% and 71%, respectively. The model accuracy hierarchy was ANN > XGBoost > RF > SVM > DT > kNN during training, and ANN/XGBoost > kNN > DT/RF > SVM during testing. This approach allows interpreters to select the most accurate ML model based on training phase performance. This study provided a clear insight towards generating a supplement for litholog sequence and improving the accuracy and efficiency of lithology prediction in a geologically complex petroleum reservoir using pre-received core derived litholog information at few wells.