Yicheng Song

Accurate determination of rock mechanical parameters for ultra-deep carbonate reservoirs (>9000 m) is critical for wellbore stability evaluation and acid-fracturing parameter optimization, yet systematic experimental data for such depths remain scarce. This study investigates grainstone cores from the YS Section (9199.8–9624.0 m) of Well XY in the Tarim Basin using continuous scratch tests—a minimally destructive method providing continuous strength profiles—complemented by conventional uniaxial compression tests. Five working conditions are systematically compared: original (unsoaked) state, drilling-fluid instantaneous contact (0 h), fresh water immersion (24 h), drilling-fluid immersion (24 h), and 15% hydrochloric-acid immersion (24 h). In the original state, scratch-test averages yield a uniaxial compressive strength (UCS) of 179.9 MPa, internal friction angle of 34.2°, and cohesion of 47.6 MPa. None of the three water-based working fluids causes measurable strength degradation within 24 h, with UCS values of 191.6, 186.0, and 190.9 MPa, respectively, attributable to the low water sensitivity of carbonate minerals. In contrast, 24 h acid immersion reduces UCS to 109.2 MPa (−39.3%), cohesion to 29.0 MPa (−39.0%), and internal friction angle to 27.1° (−20.6%), driven by intergranular cement dissolution, with strong sample-dependent variability (individual UCS reductions ranging from 12.6% to 48.0%). Scratch depth significantly influences results: UCS at 0.20 mm depth is 30.3% lower than at 0.18 mm. Calibration against rigid-testing-machine UCS yields a ratio coefficient α = 0.47~0.61. A value of 0.56 is set for α, giving a recommended engineering UCS of 100.7 MPa. These findings provide an experimental basis for wellbore stability analysis and acid-fracturing design in ultra-deep carbonate reservoirs.

The Wumishan Formation carbonate buried-hill reservoir in Hejian, Jizhong Depression, is a key geothermal reservoir where natural fractures control high-yield production. However, sparse control wells and seismic attribute non-uniqueness render conventional manual fusion subjective, while complex models tend to overfit under small-sample conditions. This study presents a well-controlled seismic multi-attribute fusion method that uses four-level (I–IV) well-log reservoir fracture interpretation as supervised calibration labels to predict 3D reservoir fracture distribution. Using 3D seismic data and comprehensive well-log fracture interpretation from wells JR1-1 and JR1-6, more than ten fracture-sensitive attributes were computed and systematically compared against well fracture grades through well-seismic calibration. Three complementary attributes—high-resolution eigenvalue coherence, maximum principal component curvature, and maximum likelihood similarity enhancement—were selected as primary fusion channels. An ordinal logistic regression model was then trained on the well-calibrated samples, jointly solving feature weights and ordinal cumulative classification thresholds within a unified likelihood framework to map seismic features onto the four-level reservoir fracture classification. Application to the target block shows that data-driven attribute weights are consistent with expert manual weights, and leave-one-well-out cross-validation yields 80.5–84.1% reproducibility (Kappa = 0.744). The resulting reservoir fracture distribution exhibits pronounced spatial zonation and vertical segmentation, with dominant fractures concentrated at structural highs on the reservoir top surface. This method provides a robust, well-controlled framework for natural fracture characterization and can guide geothermal well target optimization and reservoir stimulation design in deep geothermal fields.

The continuous prediction and uncertainty quantification of unconfined compressive strength (UCS) and the fracture-related index of anisotropy (FRIA) are essential for optimizing drilling operations and hydraulic fracturing design in shale gas development. However, machine-learning-based log inversion often suffers from (1) spatial information leakage caused by autocorrelation in well logs, (2) implicit target contamination during multi-source data fusion, and (3) biased evaluation under random data splitting, which can overestimate apparent performance and underestimate extrapolation risk in deep heterogeneous intervals. To address these limitations, we propose a leakage-free, physics-constrained framework for predicting UCS and FRIA in the Weiyuan shale gas reservoir. Using 18,440 quality-controlled, depth-aligned samples, we adopt a contiguous depth-based split that preserves stratigraphic continuity while isolating training, validation, and test intervals to block spatial leakage. Under a strict leakage-free protocol, we evaluate single-task ensemble trees (STL-RF/HGB), a multi-task neural network (MTL-MLP), and a physics-informed variant (PINN-MLP) for deep-interval stabilization. The best model is target-dependent: STL-RF achieves R² = 0.984 for FRIA, whereas MTL-MLP attains R² = 0.874 for UCS. For deep formations (>4800 m), PINN-MLP with a depth-continuity constraint reduces deep-interval prediction error by 47.5%. Multi-seed experiments with 95% Student’s t confidence intervals further confirm robustness. Overall, the framework provides a reproducible workflow for continuous geomechanical-parameter prediction and risk-aware deployment in deep unconventional reservoirs.