Quantitative Classification of Microscopic Pore Structure in Carbonate Reservoirs Using Multi-Source Data Fusion and Machine Learning Integration

Microscopic pore structure strongly controls hydrocarbon storage and flow in carbonate reservoirs, but objective and continuous pore-type classification remains difficult because carbonate pore systems are multiscale, heterogeneous, and commonly interpreted using experience-based criteria. This study develops a reproducible workflow that integrates 912 mercury-intrusion capillary pressure (MICP) datasets from 34 wells with 474 paired thin-section and core-photograph observations from the S oilfield. Principal component analysis (PCA) reduces eight pore-structure parameters to three interpretable components that describe pore-throat scale, distribution uniformity, and connectivity/displacement behavior, retaining 87.63% of the total variance. K-means clustering identifies four pore types for dolomite and four for limestone, with k = 4 selected using the elbow criterion, silhouette coefficient, centroid interpretability, and petrographic consistency. Modified injection-to-final-state analysis (MIFA) is used as an internal MICP-based consistency check rather than as a fully independent validation; paired micro-observations provide cross-scale validation with 81.22% agreement. Lithology-constrained GR, SP, and AC response windows are then used for intra-field upscaling to uncored intervals, and field-scale back-checking shows 87% agreement with existing geological interpretations. The workflow reduces interpreter subjectivity, provides physically interpretable pore-type criteria, and is applicable to carbonate reservoirs with comparable MICP, petrographic, and logging constraints.