Print this pageSuccessful Carbonate Well Log Facies Prediction Using from Artificial Neural Network Method - Wafra Maastrichtian Reservoir, Partitioned Neutral Zone (PNZ), Saudi Arabia and Kuwait
The Upper Cretaceous-age Maastrichtian reservoir is one of five major oil reservoirs in the giant Wafra oil field. The Maastrichtian oil production is largely from subtidal dolomites deposited on a very gently dipping, shallow, arid, and restricted ramp that transitioned between normal marine and restricted lagoonal conditions.
The reservoir has six cored wells, five of which are along the axis of the reservoir. Consequently, stratigraphic and depositional interpretation has been limited. This study was done to determine (1) if facies could be successfully predicted from routinely acquired well log data and (2) if the predicted facies could be used to improve the stratigraphic and depositional interpretation of the reservoir. Efforts to predict facies from well logs in carbonate reservoirs is difficult due to complex facies structures, strong diagenetic overprint, and challenging log analysis due in part to the presence of vugs and fractures.
The first step of the workflow involved an extensive input data preparation effort that involved additional well log processing, analysis, and normalization along with preparation of digital facies descriptions in which the 20 “micro-facies” categories in the original core descriptions were reduced to six significant facies categories (mudstone, packstone, mud-rich packstone, grain-rich packstone, grainstone, and floatstone), Following evaluation of several alternative algorithms, a probabilistic neural network (PNN) algorithm was used to predict facies from the well log data. The PNN algorithm successfully delineated the complex nonlinear relationships between carbonate facies and available well log data. Hold-off data from the training wells as well as blind tests on cored wells not included in the neural network training were used to validate the facies predictions. Results obtained on blind well tests show an overall prediction accuracy of more than 75%. The prediction uncertainty is quantified by two probabilistic logs, discriminant ability and overall confidence, which help the geologist to evaluate facies prediction limitations as the predicted facies logs are used to supplement core well data as part of on-going stratigraphic and depositional setting studies as well as preparation of facies distribution maps.
AAPG Search and Discovery Article #90090©2009 AAPG Annual Convention and Exhibition, Denver, Colorado, June 7-10, 2009