مشخصات پژوهش

Shale volumes are essential for lithology identification, reservoir evaluation, and stratigraphic correlation in subsurface formation analysis. This study evaluates the performance of three machine learning algorithms—Random Forest (RF), Extra Trees (ET), and Artificial Neural Network (ANN)—for shale volume prediction using conventional well log data. The models were trained and tested on a comprehensive dataset from one of the oil fields in southern Iran, incorporating parameters such as sonic travel time (DTC), bulk density (RHOZ), resistivity (RT), neutron porosity (HTNP), and caliper (HCAL). Results demonstrated that ANN achieved superior accuracy with an R² of 0.9779 and Root Mean Squared Error (RMSE) of 0.0130 API, outperforming both RF (R² = 0.9640, RMSE = 0.0166 API) and ET (R² = 0.9007, RMSE = 0.0275 API). While ANN excelled in capturing complex nonlinear relationships, tree-based methods offered faster training times and greater interpretability through feature importance metrics. The findings highlight ANN as the preferred choice for high-fidelity shale volume prediction, whereas RF provides a balanced solution for scenarios requiring both speed and competitive accuracy. This study underscores the transformative potential of machine learning in petrophysical analysis, offering practical recommendations for model selection based on project-specific needs.