[Objective] Deep shale reservoirs are characterized by high temperature, high pressure, elevated in situ stress, and strong plasticity. Conventional in situ stress testing methods and log interpretation models, often calibrated under simplified laboratory conditions, suffer from limited predictive accuracy, high operational cost, and poor generalizability—challenges that constrain their utility in guiding shale gas exploration and development. [Methods] Focusing on a structurally complex shale gas block in southern Sichuan as a representative case, we integrated dynamic and static multi-source data across drilling, logging, testing, and production stages. Along with experimental measurements of the physical and mechanical properties of rock under different conditions, we developed a hybrid stress prediction model by combining machine learning techniques with geomechanical principles. This multi-method, log-based intelligent prediction framework enables the generation of high-resolution in situ stress profiles for efficient shale gas development. [Results] In the first member of the Longmaxi Formation, organic content, confining presssure, and lamination structures significantly influence mechanical anisotropy, particularly in sublayers ①–④, which exhibit stronger anisotropy than the upper layers. Based on these findings, we developed a mechanically calibrated anisotropic stress interpretation model for deep shales. Using laboratory- and field-calibrated synthetic stress datasets, we established a standardized stress database for the southern Sichuan shale reservoir. Key sensitive logging parameters, including shear wave slowness, resistivity, acoustic logs, and Young's modulus were identified via Pearson correlation analysis. An optimized XGBoost model achieved interpretation accuracies above 90 % for all three principal stress components, with RMSE of 6.63, MAE of 3.89, and a coefficient of determination (R²) of 0.91, indicating strong robustness and generalizability. The results revealed four distinct stress barrier layers within the Wufeng–Longmaxi formations; the top of layer ① and layer ⑥ acted as dominant stress-sealing interfaces. Localized compressive stresses induced by fold-related deformation further enhanced vertical stress compartmentalization and increased the minimum horizontal principal stress, thereby exerting significant influence on hydraulic fracturing performance. [Conclusion] The study revealed that the lower sublayers of the Longyi Formation shale exhibit stronger mechanical anisotropy due to factors such as organic matter content and mineral fabric, leading to the development of an isotropic geostress interpretation model suitable for deep shale reservoirs. By constructing a data-driven intelligent prediction model (L-XGBoost), high-precision interpretation of triaxial geostress (accuracy >90 %) was achieved. It was also clarified that four sets of stress barrier layers exist within the shale, among which the strong stress barriers and local compressive stress can increase the minimum horizontal principal stress, thereby adversely affecting fracturing outcomes. [Significance] These insights provide a scientific basis for fine-scale stratigraphic subdivision and three-dimensional well pattern design for shale gas development in the tectonically complex southeastern Sichuan Basin.
DownLoad: