Probabilistic Inversion and Dynamic Identifiability Assessment of Oil/Water Contact Using Water-Breakthrough Timing in Fault-Controlled Carbonate Reservoirs

Abstract

Determining local oil/water contact (OWC) depth in fault-controlled carbonate reservoirs remains difficult because ultradeep wells rarely penetrate the full oil/water transition zone. Within a connected unit, pressure communication may coexist with different local equivalent OWC depths and water-breakthrough responses. Water-breakthrough timing provides a widely available dynamic constraint, but it is also affected by fault-zone flow capacity, interwell connectivity, aquifer support, and production conditions. Matching breakthrough timing alone, therefore, does not establish a unique local OWC. This study develops a surrogate-assisted probabilistic workflow to test whether the 10% water-cut breakthrough time, BT10, constrains local OWC depth at each well, and, where it does, to estimate a posterior OWC interval. The workflow is applied to a three-well fault-controlled carbonate connected unit in an ultradeep Ordovician reservoir in the Tarim Basin, northwest China. Three local equivalent OWC depths are treated as inversion targets, while damage-zone permeability enhancement, W5–W6 transmissibility, and aquifer-support strength are treated as uncertain controls. A library of 1,000 simulations is generated with a 3D black-oil model implemented in INTERSECT (SLB, Houston, Texas, USA). Well-specific extreme gradient boosting (XGBoost) surrogates are trained on 800 cases and evaluated on 200 held-out cases. The surrogates then evaluate 50,000 candidates within the same prior ranges. The surrogates achieve normalized mean absolute errors (MAEs) of 2.1%, 4.7%, and 3.1% for W4, W5, and W6, respectively, and provide a per-evaluation speedup of approximately 1.6×105. In three representative posterior recomputations, the surrogate/simulator difference in BT10 does not exceed 9 days. The deterministic reference case reproduces W4 breakthrough but predicts breakthrough 46.2% and 41.8% too early at W5 and W6. At W4, the P10–P90 OWC interval narrows from 112.8 m to 35.3 m, reducing the completion-base-to-contact clearance range from 78.3–191.1 m to 115.4–150.7 m. At W5 and W6, compensation among local OWC depth, connectivity, and aquifer support leaves the posterior OWC intervals essentially unchanged. Single-well conditioning reproduces the same identifiability ranking, showing that the contrast is not created by crosswell weighting. BT10 is therefore not a universal contact-depth indicator. The workflow tests whether the available breakthrough data constrain local OWC depth. Where they do not, the analysis indicates the modeled control most strongly associated with nonuniqueness within the tested parameter space and the corresponding diagnostic need.