Using Neural Network Predicted Facies to Refine the Stratigraphic Framework; Maastrichtian Reservoir from Wafra Field, Partitioned Neutral Zone (PNZ), Saudi Arabia and Kuwait
The Maastrichtian (Upper Cretaceous) reservoir at Wafra Field is targeted for increased production by Saudi Arabia Chevron (SAC). The Maastrichtian sediments are dominated by subtidal dolomites deposited on a shallow, gently dipping ramp. Lithofacies are dominantly wackestone, rudist rudstone, peloidal packstone, and localized grainstone, with minor mudstone and rudist floatstone. Moldic, inter- and intra-crystalline porosity are common. The average porosity is approximately 15% and the average permeability is about 30mD.
Previous stratigraphic studies have relied on limited core data and wireline logs to develop a field wide and regional stratigraphic framework. In this study, a probabilistic neural network was used to accurately predict facies from well log data. The neural network approach has the ability to delineate complex nonlinear relationships between facies and log data. Core descriptions and log data from four Maastrichtian wells were used as training sets and allowed us to develop predicted facies strip logs with accuracy above 70%. These facies strip logs were developed for 15 key wells and then used to validate and refine the stratigraphic architecture.
The predicted facies strip logs effectively act as additional cores. However, the probabilistic neural network analysis results in numerous thin and rapidly alternating layers, which are not recognized in surrounding cores. These results must be smoothed or lumped before attempting to correlate to core data. By building cross sections that incorporate existing core data and predicted facies data, a more consistent, field wide stratigraphic framework is developed.
AAPG Search and Discovery Article #90090©2009 AAPG Annual Convention and Exhibition, Denver, Colorado, June 7-10, 2009