Evaluating Different Approaches to Permeability Prediction in a Carbonate Reservoir

Abstract

Permeability can be directly measured using cores taken from the reservoir in the laboratory. Due to high cost associated with coring, cores are available in a limited number of wells in a field. Many empirical models, statistical methods, and intelligent techniques were suggested to predict permeability in un-cored wells from easy-to-obtain and frequent data such as wireline logs. The main objective of this study is to assess different approaches to the prediction of the estimation of permeability in a heterogeneous carbonate reservoir, i.e. Fahliyan formation in the southwest of Iran. The considered methods may be categorized in four groups, namely a) empirical models (Timur and Dual-Water), b) regression analysis (simple and multiple), c) clustering methods like MRGC (multi-resolution graph-based clustering), SOM (self organizing map), DC (dynamic clustering) and AHC (ascending hierarchical clustering), and d) artificial intelligence techniques such as ANN (artificial neural network), fuzzy logic, and neuro-fuzzy. This study shows that clustering techniques predict permeability in a heterogeneous carbonate better than other examined approaches. Among four assessed clustering methods, SOM performed better and correctly predicted local variations. Artificial intelligence techniques are average in modeling permeability. However, empirical equations and regression methods are not capable of predicting permeability in the studied reservoir. The constructed and validated SOM model with 6×9 clusters was selected to predict permeability in the blind test well of the studied field. In this well, the predicted permeability was in good agreement with MDT and core derived permeability.