This study addresses the persistent challenges of ambiguous oil-water relationships and poorly defined effective reservoir distribution in high-mud-content sandstone reservoirs of the III Oil Group (Zhujiang Formation, Member 1) within the Wenchang X-3 structure. Utilizing high-precision reprocessed 3D seismic data, we integrate detailed well-seismic calibration with sedimentary models to train a deep learning algorithm for the automatic identification of seismic facies characterized by “tidal sand” reflections. This approach enables the rapid and quantitative mapping of three distinct phases of sand, clarifying their boundaries, stacking patterns, and overall sedimentary architecture. Focusing on the prediction of effective reservoirs, the analysis begins with rock particle support and mineral composition. A dry rock frame model is established based on the Hashin-Shtrikman (HS) bounds, deriving quantitative relationships between key elastic parameters (e.g., P- and S-wave impedance) and reservoir properties including mud content, porosity, and oil saturation. This work culminates in the first development of a specialized rock physics template for marine high-mud-content sandstone in this region. To overcome the lack of diagnostic elastic responses, dozens of seismic and geological attributes were screened and optimized using machine learning. An innovative sensitive factors of reservoirs was constructed based on the Vp/Vs ratio and the P-wave impedance (Ip), which field application confirms to be highly accurate for identifying effective reservoirs. This research is the first to reveal a three-phase, superimposed tidal sand system in the Shenhu Uplift, explaining its planar distribution, migration history, and thereby resolving prior oil-water contradictions. The proposed predictor effectively reduces the interpretative complexity of high-mud-content sandstone reservoirs, clearly delineates their 3D spatial distribution, and provides a reliable technical foundation for validating this category of lithologic traps.
Wang et al. (Wed,) studied this question.