Determination and prediction of Pore-facies based on a Compilation of Mercury Injection, Petrophysical and Petrographic Data Using a Hybrid of Self-organizing Neural Network and Support Vector Machine

Pore network characteristics control the fluid condition in reservoir rocks. In carbonate reservoirs, fluid flow status is independent of primary depositional texture, so network properties must be directly included in the process of facies determination to accomplish them and make them be applicable for analyzing the reservoir real conduct. A compilation of petrographic, petrophysical and reservoir engineering studies is carried out to characterize pore throats and lithofacies using Self-organizing Map Neural Network in Dalan and Kangan formations of South Pars Gas Field in this paper. Five pore-facies with unique petrophysical, geological and reservoir features are determined by the applied network. A sharp decreasing trend in reservoir quality recognized from pore-facies 1 toward 5 based on their extracted properties. Meanwhile, Support Vector Machine (SVM) which was used for prediction of pore-facies identified in previous steps from wireline well logs. The accuracy of the model in prediction of pore-facies is 78% which indicates an acceptable result for the model in the South Pars Gas Field.