GANSim: Conditional geological modelling with Generative Adversarial Networks (GANs)

Suihong Song
Facies geomodelling is significant for evaluation and exploitation of subsurface earth resources as well as for geological sequestration of CO2 (CCS). Based on GANs, we propose GANSim framework to directly produce multiple realistic geomodels, conditioned to given global features (e.g., facies proportion, channel width), sparse well data, and geophysics-produced facies probability maps. GANSim has been validated with synthetic channelized reservoirs, and has been practically used for geomodelling of 3D karst caves, inner architecture of point bars, 3D deltas, and 3D turbidites. It has also been combined with physics-informed CNN surrogate to further achieve conditioning of geomodels to well flow data.