Diagenetic facies characteristics and quantitative prediction via wireline logs based on machine learning: A case of Lianggaoshan tight sandstone, fuling area, Southeastern Sichuan Basin, Southwest China

Tight sandstone has low porosity and permeability, a complex pore structure, strong heterogeneity due to diagenetic modifications. Limited intervals of Lianggaoshan Formation in the Fuling area are cored high costs, thus, model for predicting facies based on logging curves was established few core, thin section, X-ray diffraction (XRD), scanning electron microscopy (SEM), cathodoluminescence, routine core analysis, mercury injection capillary pressure tests. The results show that tight primary secondary intergranular pores, intragranular micropores clay minerals. compaction experienced by is medium strong, main minerals carbonates (calcite, dolomite, ferric dolomite) (chlorite, illite, mixed illite/montmorillonite). Four types recognized: carbonate cemented (CCF), tightly compacted (TCF), chlorite coating mineral filling (CCCMFF), dissolution (DF). Primary pores develop CCCMFF, DF; porosities permeabilities CCCMFF DF better than CCF TCF. were converted data, prediction using four machine learning methods established. random forest highest accuracy 97.5%, followed back propagation neural networks (BPNN), decision trees, K-Nearest Neighbor (KNN). In addition, had smallest difference between different (2.86%). Compared with other three models, can balance unbalanced sample data improve area, which wide application range. It worth noting BPNN may be more advantageous when there types.