Prediction of complex diagenesis and the resulting variations in reservoir quality are critical for hydrocarbon exploration. In fact, the limitations of core accessibility and expensive costs of core-based experiments have posed challenges for establishment of diagenetic prediction model. Integrated petrographic and petrophysical analyses are served here to fully understand the dominant diagenetic features and their controls on reservoir quality. Six diagenetic facies representing distinct mineralogical compositions, diagenetic processes, and pore systems were then categorized. To upscale diagenetic features through correlating core diagenetic facies with geophysical well logs, linear discriminant analysis (LDA) was first employed to obtain eigenvectors that can best describe and distinguish diagenetic facies. The well logs that have stronger influence on the first and second eigenvectors and have less correlations with each other are selected to compress the feature space dimensions. The supervised self-organizing map (SSOM) predicable model was trained using dimensionally reduced well log database as input and core diagenetic facies as supervision to determine the nonlinear mapping relations between log response combination features and diagenetic facies group membership. The results showed that LDA-assisted SSOM model yielded higher accuracy as compared with commonly employed linear and non-linear predictable model. The supervised LDA-assisted SSOM method as provided here can be gainfully used to predict diagenetic facies via conventional well logs.
Read full abstract