An Improved Method of Reservoir Facies Modeling Based on Generative Adversarial Networks

As the reservoir and its attribute distribution are obviously controlled by sedimentary facies, facies modeling is one of important bases for delineating area high-quality characterizing parameter distribution. There a large number continental reservoirs with strong heterogeneity in China, geometry various microfacies relatively complex. The traditional geostatistics methods which have shortage characterization complex non-stationary geological patterns, limitation reservoirs. generative adversarial network (GANs) recent state-of-the-art deep learning method, has capabilities pattern generation, widely used domain image generation. Because similarity content structure between models specific images (such as fluvial modern rivers), generated GANs often more than models, potential to be modeling. Therefore, this paper proposes method based on GANs: (1) unconditional modeling, select training (TIs) priori knowledge, use learn patterns TIs, then generate model GANs; (2) conditional “unconditional-conditional simulation cooperation” (UCSC) realize constraint hard data while patterns. Testing using both synthetic actual from oil field, results meet perfectly honor well point data, show that can overcome difficult deal improve effect methods. Given good performance prospect practical application.