--> ABSTRACT: Emerging Technologies in Reservoir Prediction: Application of Neural Networks to Siliciclastic and Carbonate Reservoir Definition, by P. Crevello, M. T. Binfi, and M. Prins; #91021 (2010)
[First Hit]

Datapages, Inc.Print this page

Emerging Technologies in Reservoir Previous HitPredictionNext Hit: Application of Neural Networks to Siliciclastic and Carbonate Reservoir Definition

CREVELLO, PAUL, MINFI THAN BINFI,  and PRINS, MAX

Neural network modeling of lithofacies, porosity and permeability is an emerging technology in the characterization of siliciclastic and carbonate reservoirs, especially insituations when conventional methods yield equivocal results. The methodology involves training the network with core lithofacies and petrophysical data, and wireline well-log data. The network is tested on untrained, cored intervals and applied to uncored wells.

Network success is related to the complexity of the reservoir: i.e., simple or complex lithofacies zonation, thick vs. thin bed, porosity and permeability variation and types, and variability of fluid content. In complex reservoir systems, i.e., multiple lithofacies and reservoir zonations, subtle lithofacies may be below petrophysical and wireline discrimination, such that Previous HitpredictionNext Hit is non-unique. The predictive success of lithofacies in complex networks ranges between 40-88%, but with an overall success of 74%. Similar lithofacies with minor depositional/petrophysical distinctions will have the lowest success rate. Simplifying the reservoir into fewer broadly related lithofacies improves the Previous HitpredictionNext Hit. This is evident in a simple network which has 4 reservoir zones: the result is a 95% success in lithofacies Previous HitpredictionNext Hit, which also facilitate reservoir modeling studies.

In thick bedded reservoir sequences, network porosity and permeability calculations are reasonably accurate and within acceptable tolerances used in reservoir studies. Previous HitPredictionNext Hit of permeability is generally unreliable in complexly bedded sequences of thin-beds or variable HC saturations, failing by an order of magnitude in intermediate permeability ranges (100-1000md).

Overall, the application of neural networks to lithofacies, porosity, and permeability analyses proved highly successful in the Previous HitpredictionTop of siliciclastic and carbonate reservoirs and aids in zonation away from cored wells. This methodology is rapidly gaining popularity as reservoir characterization requires multidisciplinary evaluation. 

AAPG Search and Discovery Article #91021©1997 AAPG Annual Convention, Dallas, Texas.