Conditioning Stratigraphic, Rule-Based Models With Generative Adversarial Networks: A Deepwater Lobe Example
A stratigraphic, rule-based reservoir modeling method approximates sedimentary dynamics to generate numerical descriptions of reservoir architecture while capturing geological realism. However, robust conditioning these models to local data (i.e., well log interpretation, core data, and seismic constraints) is a remaining challenge. This study suggests an opportunity for novel deep learning-based method for the fast and flexible condition of local data. We design a rule-based modeling method for a deep-water lobe reservoir, controlled by three geological parameters: 1) the compositional exponent, 2) the lobe geometry, and 3) the distribution of petrophysical properties within the lobes. The compositional exponent tunes the placement rule of lobe elements based on the elevation of the previous composite surface. The geometries of the lobe elements are determined by their radii and thicknesses. After building the reservoir structure, the rule-based method allocates petrophysical properties (i.e., porosity and permeability) with a hierarchical trend model, parameterized by a mean and a standard deviation. Generative Adversarial Networks (GAN) uses the multiple realizations of the rule-based model to learn to extract the main geological features and generate reservoir models that preserve the features. With the trained model, we can navigate in the latent reservoir model manifold and find the optimum models that satisfy the given local data. An experiment on a typical deep-water depositional complex demonstrates that our approach successfully captures the geological rule and the induced features. Moreover, we can preserve local data while rendering realistic reservoir heterogeneity, continuity, and spatial distribution of petrophysical parameters. The flexibility of the rule-based modeling method and GAN enables our approach to be applied to various depositional systems.
AAPG Datapages/Search and Discovery Article #90350 © 2019 AAPG Annual Convention and Exhibition, San Antonio, Texas, May 19-22, 2019