Parametric generation of conditional geological realizations using generative neural networks

摘要: Deep learning techniques are increasingly being considered for geological applications where -- much like in computer vision the challenges characterized by high-dimensional spatial data dominated multipoint statistics. In particular, a novel technique called generative adversarial networks has been recently studied parametrization and synthesis, obtaining very impressive results that at least qualitatively competitive with previous methods. The method obtains neural network of geology so-called generator is capable reproducing complex patterns dimensionality reduction several orders magnitude. Subsequent works have addressed conditioning task, i.e. using to generate realizations honoring observations (hard data). current approaches, however, do not provide conditional generation process. this work, we propose obtain direct realizations. main idea simply extend existing stacking second inference learns perform conditioning. This trained sample posterior distribution derived Bayesian formulation task. resulting extended thus provides parametrization. Our assessed on benchmark image binary channelized subsurface, promising wide variety configurations.