3D CNN-PCA: A deep-learning-based parameterization for complex geomodels

Geological parameterization enables the representation of geomodels in terms of a relatively small set of variables. Parameterization is therefore very useful in the context of data assimilation and uncertainty quantification. In this study, a deep-learning-based geological parameterization algorithm, CNN-PCA, is developed for complex 3D geomodels. CNN-PCA entails the use of convolutional neural networks as a postprocessor for the low dimensional principal component analysis representation of a geomodel. The 3D treatments presented here differ somewhat from those used in the 2D CNN-PCA procedure. Specifically, we introduce a new supervised learning-based reconstruction loss, which is used in combination with style loss and hard data loss. The style loss uses features extracted from a 3D CNN pretrained for video classification. The 3D CNN-PCA algorithm is applied for the generation of conditional 3D realizations, defined on 60 x 60 x 40 grids, for three geological scenarios (binary and bimodal channelized systems, and a three-facies channel-levee-mud system). CNN-PCA realizations are shown to exhibit geological features that are visually consistent with reference models generated using object-based methods. Statistics of flow responses (P-10, P-50, P-90 percentile results) for test sets of 3D CNN-PCA models are shown to be in consistent agreement with those from reference geomodels. Lastly, CNNPCA is successfully applied for history matching with ESMDA for the bimodal channelized system.

Automated summary

Geological parameterization allows for representation of geomodels with a small set of variables, useful for data assimilation and uncertainty quantification. A deep-learning-based CNN-PCA algorithm is developed for complex 3D geomodels, using convolutional neural networks and principal component analysis. The algorithm is applied successfully for generating conditional 3D realizations for different geological scenarios, showing consistent agreement with reference models.