Unconventional oil and gas reservoirs exhibit strong heterogeneity, complex hydrocarbon accumulation mechanisms, and low accuracy with high computational costs in traditional numerical simulations. This paper proposes an integrated workflow combining reservoir sedimentary feature-driven numerical simulation of hydrocarbon accumulation mechanisms with intelligent identification algorithms. Taking the Ji Yuan Chang 6 reservoir in the Ordos Basin as a case study, this approach integrates multi-source data including 3D seismic, well logging, and core samples to construct multi-point geostatistical (MPS) training images. This enables automated sedimentary facies classification and high-precision reconstruction of 3D heterogeneous models, improving drilling correlation rates from 78.5% to 92.3% and enhancing sand body connectivity by over 50%. Building upon this foundation, a coupled black oil-component hybrid numerical simulator for sedimentary facies-controlled parameter fields was developed. Capillary pressures, relative permeabilities, and unconformity conductivity properties of different microfacies were explicitly incorporated into the flow equations. Utilizing adaptive mesh refinement, single-run computation time was reduced from >72 hours to <8 hours. The historical fit error decreased from 22.5% to 8.2%, accurately reproducing the enrichment patterns of overthrust-type “isolated” reservoirs. Further, a knowledge-data dual-driven graph neural network (GNN) was designed. Expert rules—such as hydrostatic pressure and structural high accumulation— were converted into differentiable loss constraints, addressing overfitting in small samples. The mean absolute error for predicting oil saturation in blind wells decreased to 0.081, with the coefficient of determination rising to 0.89. The geological plausibility score significantly outperformed traditional GNNs. This research achieves a closed-loop system integrating sedimentary control, fluid migration, and intelligent identification, providing a scalable new paradigm for efficient exploration and development of deep low-permeability and unconventional oil and gas reservoirs.
Yingqi Song (Sun,) studied this question.