Nuclear Magnetic Resonance–Based Quantitative Evaluation of Shale Oil Reservoir Fluid Types Using Numerical Inversion and a Prior-Constrained Bayesian-Gaussian Mixture Model

Summary Continental shale oil reservoirs have diverse pore sizes and significant heterogeneity, making fluid discrimination and saturation quantification challenging with conventional 2D nuclear magnetic resonance (NMR) logging. In this study, we propose a method for multiphase fluid identification, combining numerical simulation, core-scale experiments, and a prior-constrained Bayesian-Gaussian mixture model (PC-BGMM). Pore-scale geometries are extracted from high-resolution scanning electron microscopy images, and multiphase NMR responses are forward-modeled by incorporating bulk and surface relaxation mechanisms. The simulated results are calibrated and validated using deuterium oxide (D2O) suppression and temperature-gradient drying experiments, enabling the construction of a comprehensive 2D NMR fluid identification scheme. Prior physical information is then embedded into the PC-BGMM framework to achieve unsupervised classification of multiphase fluids and continuous inversion of fluid saturations. Application to the Lianggaoshan Shale Oil Reservoir demonstrates that the proposed method enables stable discrimination of multiple fluid components from sample to well scales. The inverted saturations of movable oil and free gas show strong consistency with production test data and effectively characterize productivity variations among different reservoir types.