Synthetic Slowness Shear Well-Log Prediction Using Supervised Machine Learning Models.

The shear slowness well-log is a fundamental feature used in reservoir modeling, geomechanics, elastic properties, and borehole stability. This data is indirectly measured by well-logs and assists the geological, petrophysical, and geophysical subsurface characterization. However, the acquisition of shear slowness is not a standard procedure in the well-logging program, especially in mature fields that have a limited logging scope. In this research, we propose to develop machine learning models to create synthetic shear slowness well-logs to fill this gap. We used standard well-log features such as natural gamma-ray, density log, neutron porosity, resistivity logs, and compressional slowness as input data to train the models, and successfully predicted a synthetic shear slowness well-log. Additionally, we created five supervised models using Neural Networks, AdaBoost, XGBoost, and CatBoost algorithms. Among all models created, the neural network algorithm provided the most optimized model, using multi-layer perceptron architecture reaching impressive scores as R $$^2$$ of 0.9306, adjusted R $$^2$$ of 0.9304, and MSE less than 0.0694.