Multiphase fluid identification in the Lianggaoshan formation shale-oil reservoir by integrating T1-T2 NMR logging and numerical simulation

• A surface–bulk relaxation model, calibrated with SEM‑derived pore geometry, is used to forward‑simulate 2D NMR (T₁–T₂) responses of seven fluid occurrences in shale. • A new free‑gas domain (T₁ > 300 ms, T₂ > 300 ms, T₁≈ T ₂) is introduced, extending conventional shale‑fluid atlases. • Prior windows from simulation are embedded in a constrained K‑means routine that automatically partitions field spectra and yields continuous, seven‑phase saturation curves. • Movable‑oil and free‑gas amplitudes predicted by the workflow correlate with production‑test data, whereas high‑porosity layers dominated by bound fluids remain non‑productive. • The method quantifies fluid mobility without pore‑size transforms or extra core tests, offering a rapid, transferable tool for unconventional‑reservoir NMR interpretation. Continental shale-oil reservoirs present interpretation challenges because of their wide pore-size spectrum, multiphase fluids, and the frequent presence of free gas-features that conventional two-dimensional nuclear-magnetic-resonance (2D-NMR) logging rarely resolves. To remedy this gap, we target the Lianggaoshan Formation (Sichuan Basin) and aim to build a physics-informed workflow that reliably distinguishes all major fluid phases, including free gas, and quantifies their saturations without core-dependent calibrations. Materials and methods combine (i) scanning-electron-microscopy–derived pore geometries, used in a surface-to-volume relaxation model (diffusion relaxation neglected), with (ii) forward numerical simulation that maps seven representative fluids—organic matter, clay-bound water, capillary-bound water, bound oil, movable oil, movable water, and free gas—onto the T₁–T₂ plane. The resulting domains, including a newly defined free-gas window (T₁ > 500 ms, T₂ > 500 ms, T₁ ≈ T ₂), serve as prior constraints in a K-Means algorithm that automatically zones 2D-NMR logs. Application to Well G119H (27 fracture stages, 2740–4040 m) demonstrates that layer-scale variations in movable-oil and free-gas amplitudes track production-test statistics, whereas high-porosity but bound-fluid-dominated layers remain non-productive. Results and discussion confirm that the workflow differentiates seven fluid types and delivers continuous saturation curves without invoking empirical pore-size/T₂ transforms. The free-gas window, absent from earlier atlases, proves essential for matching gas-drive contributions observed in the production data. The approach provides a rapid, core-light tool for sweet-spot evaluation in unconventional plays; its prior domains can readily feed semi-supervised inversions or Transformer-VAE models, while forthcoming high-temperature-pressure NMR and micro-CT integration will refine bound-fluid quantification under true reservoir conditions.