Quantitative Brittleness Evaluation of Tight Sandstone in the Yanchang Formation, Southern Ordos Basin, Based on Machine Learning and Multi-Source Data

The southern Ordos Basin is rich in tight oil resources. However, the strong heterogeneity of its main reservoir, the tight sandstone of the Yanchang Formation, limits the applicability of traditional brittleness evaluation methods based on single parameters or linear combinations, affecting the accurate identification of fracturing “sweet spots”. To more authentically characterize reservoir brittleness, this study focuses on the tight sandstone of the Yanchang Formation in the southern Ordos Basin. Through high-temperature and high-pressure triaxial mechanical tests, Brazilian splitting tensile strength tests, and X-ray diffraction analysis, the controlling mechanisms of mineral composition, temperature-pressure conditions, and laminated structures on rock mechanical properties were systematically revealed. Furthermore, fractal geometry theory was introduced, and the fractal dimension of post-fracture rock surfaces was used as an index to quantitatively characterize the true brittleness. The Gradient Boosting Decision Tree (GBDT) machine learning algorithm was employed to integrate four fundamental brittleness indices based on energy, elasticity, strength, and shear mechanisms, constructing a nonlinear comprehensive brittleness index (CBI) prediction model. The results indicate that: (1) The fracture fractal dimension can effectively quantify the complexity of rock failure and shows a regular decrease with increasing confining pressure; (2) The GBDT model can effectively capture the complex nonlinear relationships among multi-source parameters, achieving a coefficient of determination (R²) of 0.9749 for the full sample set and a cross-validation root mean square error (RMSE) of 0.0467; (3) Feature importance analysis reveals that the contribution of the strength parameter (the ratio of tensile to compressive strength, 0.4584) is significantly higher than that of other parameters, indicating that the “tensile-compressive strength difference” is the dominant mechanical mechanism controlling brittleness development in the tight sandstone of this area; (4) By applying a dynamic rock mechanical parameter interpretation model established from well-logging data to the Chang-8 Member of Well HH26 in Honghe Oilfield on the southern margin of the Ordos Basin, layers with higher CBI (0.70) exhibited a significantly lower on-site fracture pressure gradient (1.83 MPa/100 m) compared to layers with lower CBI (0.51, 2.11 MPa/100 m), and the predicted favorable brittleness zones showed good agreement with actual fracturing responses. This study establishes a quantitative brittleness evaluation method for tight sandstone, using fractal dimension as the target and machine learning as the means, clarifying the dominant role of strength parameters in brittleness evaluation. It can provide a quantitative reference for the optimization of fracturing intervals and efficient stimulation in tight oil reservoirs in the southern Ordos Basin.