Machine learning identification of diagenetic facies in tight sandstones based on sandbody classification: The Fourth Member of the Triassic Xujiahe Formation, Jianyang area, Sichuan Basin

Within tight sandstones, diagenetic features exhibit strong heterogeneity and intra-subfacies similarities, meaning that different rock units often share comparable mineral compositions and pore structures. These similarities result in overlapping petrophysical signatures that obscure distinct logging responses, thereby significantly reducing the reliability of conventional well-logging interpretations for predicting diagenetic facies and favorable reservoirs. To address the limitations of manual interpretation under the conditions of limited data, this study developed an integrated framework that combines sandbody classification, diagenetic events, and machine learning. The main conclusions are as follows: (1) Sedimentological analyses of field outcrops and cores reveal that the tight sandstones in the braided delta-front subfacies of the Fourth Member of the Triassic Xujiahe Formation can be divided into four sandbody types: main subaqueous channels (MSC), branch subaqueous channels (BSC), terminal subaqueous channels (TSC), and distal bars and sheet sands (DB&SS). (2) Based on the quantitative characterization of diagenetic intensity, four representative diagenetic facies were identified: moderate compaction and moderate dissolution (Type I), strong compaction and weak dissolution (Type II), moderate compaction and moderate cementation (Type III), and strong compaction (Type IV). (3)Using the sedimentology-diagenesis coupling framework, a one-dimensional convolutional neural network algorithm achieved a prediction accuracy of an 81.17% for diagenetic facies through multi-parameter integration (e.g., CAL × GR, AC × DEN) and a class consolidation strategy that combines Types III and IV. (4) The internal architecture of sandbodies plays a key role in controlling the distribution of diagenetic facies and reservoir quality by regulating acidic fluid activity and cement precipitation. Favorable reservoirs (Types I and II) mainly occur in the middle-to-lower intervals of the MSC and BSC. This distribution is associated with coarse-grained, quartz-rich sandstones that promote the migration of acidic fluids and the dissolution of feldspar and rock fragments. In contrast, late-stage precipitation of authigenic carbonate and quartz results in ineffective reservoirs (Types III and IV) along the margins of the sandbodies.