Subsurface Characterization using Bayesian Deep Generative Prior-based Inverse Modeling for Utah FORGE Enhance Geothermal System

Characterization of geologic heterogeneity at an enhanced geothermal system (EGS) is crucial for cost-effective stimulation planning and reliable heat production.With recent advances in computational power and sensor technology, large-scale fine-resolution simulation of coupled thermal-hydraulic-mechanical (THM) processes have been available.However, traditional large-scale inversion approaches have limited utility for the sites with complex subsurface structures unless one can afford high, often computationally prohibitive, computations.Key computational burdens are predominantly associated with a number of large-scale coupled numerical simulations and large dense matrix multiplications derived from fine discretization of the field site domain and a large number of THM and chemical (THMC) measurements.In this work, we present deep-generative model-based Bayesian inversion methods for the computationally efficient and accurate characterization of EGS sites.Deep generative models are used to learn the approximate subsurface property (e.g., permeability, thermal conductivity and elastic rock properties) distribution from multipoint geostatistics-derived training images or discrete fracture network models as a prior and accelerated stochastic inversion is performed on the low-dimensional latent space in a Bayesian framework.Numerical examples with synthetic permeability fields with fracture inclusions with THM data sets based on Utah FORGE geothermal site will be presented to test the accuracy, speed, and uncertainty quantification capability of our proposed joint data inversion method.