Prediction on rock strength by mineral composition from machine learning of ECS logs

Rock strength evaluation is critical in oil and gas exploration, but traditional methods, such as empirical formulas, laboratory tests, and numerical simulations, often struggle with accuracy, generalizability, and alignment with field conditions. This study proposes the use of Random Forest and Transformer algorithms to predict rock strength from Elemental Capture Spectroscopy (ECS) logs. By utilizing the dry weight of minerals as input, the model predicts key mechanical properties, including Young's modulus, Poisson's ratio, bulk modulus, shear modulus, and uniaxial compressive strength. The findings demonstrate that mineral compositions, such as clay, quartz-feldspar-mica, carbonate, anhydrite, and pyrite, significantly influence rock strength. Specifically, clay content impacts Young's modulus, bulk modulus, and shear modulus, while quartz-feldspar-mica affects Poisson's ratio, and anhydrite is the primary factor influencing compressive strength. Positive correlations were observed between rock strength and the dry weight of anhydrite and carbonate minerals, while negative correlations emerged with clay, pyrite, and quartz-feldspar-mica. The Random Forest model outperformed the Transformer model in terms of predictive accuracy and computational efficiency. Its training time is only one three hundredth of the latter and its prediction time is just one tenth of the later, making it highly suitable for well-logging interpretation. Although the Transformer model was less computationally efficient, it exhibited strengths in predicting subsurface strength parameters, particularly in capturing spatial variations and forecasting these parameters across different spatial locations. This study introduces a novel AI-driven approach to rock strength evaluation, bridging the gap between mineral composition and mechanical properties, with significant implications for resource extraction and reservoir management.