--> Abstract: Successful Application of Neural Network Technique for Rock Typing Marrat Carbonate Reservoirs of West Kuwait, by Naveen Kumar Verma, Fahed Al-Medhadi, Christian Perrin, Rasha Al-Moraikhi, Eman Al-Mayyas, and D. Natarajan; #90077 (2008)

Datapages, Inc.Print this page

Successful Application of Neural Network Technique for Rock Typing Marrat Carbonate Reservoirs of West Kuwait

Naveen Kumar Verma1*, Fahed Al-Medhadi1, Christian Perrin2, Rasha Al-Moraikhi1, Eman Al-Mayyas1, and D. Natarajan1
2Schlumberger, Kuwait
*[email protected]

Application of the neural-network technique using electro-logs for rock-typing carbonates is challenging because the initial depositional fabric is often overprinted with diagenesis and tectonics. This presentation shows the first successful application of the neural-network technique for rock-typing the Marrat carbonate reservoirs in two producing oil fields in Kuwait. Conventional electro-logs from ten key wells were used for the rock-typing, whereas unconventional nuclear magnetic resonance (NMR) and borehole image logs from four of these wells were integrated to understand and quantify permeability enhancements.

Core descriptions identified nine litho-facies: (1) calcareous-shale, (2) argillaceous-limestone, (3–7) limestones (mudstone, wackestone, packstone, grainstone and oolitic grainstone), (8) dolomitic-limestone, and (9) anhydrite. A quality check using log data (gamma-ray, neutron and density) was performed for depth matching, environmental correction and normalization. The log response of lithofacies was analyzed where good correspondence was observed for eight electro-facies. This provided ‘log signatures’ that were used to identify the key zones and to train the neural network model. Validation of the electro-facies was performed using core lithofacies from eight wells.

In addition, thin-section micro-facies, φ/K and mercury injection capillary pressure (MICP) data from three wells were used to qualify the electro-facies as ‘rock types’. The results were consistent with the sequence stratigraphic framework. The neural network identified rock-types satisfactorily for the cored and non-cored intervals, including a blind test cored well and a non-cored well. Furthermore, unconventional logs were integrated where NMR helped to quantify vugs and fractures, while, image logs helped to differentiate fractured from non-fractured wackestones. Rock-types were derived for all non-cored wells and propagated through seismic control to build a fine-resolution geological model in one field. The model is expected to facilitate better management of pilot water-injection under implementation besides assisting further field development.


AAPG Search and Discovery Article #90077©2008 GEO 2008 Middle East Conference and Exhibition, Manama, Bahrain