Search

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

Based on research on the response mechanism of formation and reservoir response to logging curves, 12 logging curves were selected in combination with formation depth characteristics, and 4 algorithms were used to identify the formation and reservoir: logistic regression (LR), support vector machine (SVM), random forest (RF), and XGBoost. In the study block, 57 wells out of 60 wells were selected for training and learning, and the remaining three wells were used as prediction samples. The recognition of formation thickness and reservoirs is performed by each of these four machine learning algorithms, and predictive knowledge is obtained separately. It was found that the accuracy of the four algorithms for formation thickness and reservoir layer identification reached over 90%, but the XGBoost algorithm was found to be the best in terms of the four scoring criteria of F1-score, precision, recall, and accuracy. The accuracy of formation thickness identification could reach over 95%, and the correlation analysis between the logging curve and formation thickness could be performed on this basis. The results show that RMN, RLLD, and RLLS have the most obvious response to the sandstone layer, off-surface reservoir, and effective thickness layer, while CAL has the least effect on formation and reservoir identification, which can provide an effective reference for the selection and downscaling of subsequent logging curves.