Assessment of Clustering Methods for Predicting Permeability in a Heterogeneous Carbonate Reservoir

Permeability, the ability of rocks to flow hydrocarbons, is directly determined from core. Due to high cost associated with coring, many techniques have been suggested to predict permeability from the easy-to-obtain and frequent properties of reservoirs such as log derived porosity. This study was carried out to put clustering methods (dynamic clustering (DC), ascending hierarchical clustering (AHC) self organizing map (SOM) and multi-resolution graph-based clustering (MRGC)) into practice in order to predict the permeability of a heterogeneous carbonate reservoir in southwest of Iran. In addition, the results are compared with three conventional approaches, empirical models, regression analysis, and ANN. The performance of all the examined methods was compared in order to choose the best approach for predicting permeability in un-cored wells of the studied field. For all clustering methods, selecting the optimal number of clusters is the most important task. The optimal values for the number of clusters are selected by iteration. The optimal number of clusters for MRGC, SOM, DC, and AHC are 7, 9, 9, and 8 respectively. Empirical equations and regression analysis weakly predict permeability and the value of R2 parameters of both approaches are around 0.6. Generally the performance of clustering techniques is acceptable in Fahliyan formation. These techniques predict permeability between 1 and 1000 mD very well and just overestimate permeability values less than 1 mD. SOM performed the best among all examined techniques (R2=0.7911). The constructed and validated SOM model with 9 clusters was selected to predict permeability in one of un-cored wells of the studied field. In this well, the predicted permeability was in good agreement with MDT derived permeability.