DATA-SPACE INVERSION FOR PREDICTION OF FAULT SLIP TENDENCY IN CO2 STORAGE

Accurately assessing the potential for fault slip is essential in large-scale CO2 storage operations. Traditional model-based history matching methods, which entail the generation of posterior geological models calibrated to observed data, can be challenging and time-consuming to apply in coupled flow-geomechanics problems with faults. In this work, we implement a variational autoencoder (VAE)-based data-space inversion (DSI) framework to predict pressure, stress and strain fields, along with fault slip tendency, in CO2 storage projects. The DSI framework directly infers the posterior distributions of quantities of interest from observed data, eliminating the need to generate posterior geomodels. The geomodel in this study involves a 3D system with two faults. Heterogeneous permeability and porosity realizations of the storage aquifer are generated using geostatistical software, and geomechanical and fault parameters are sampled for each realization from predefined prior distributions. Coupled flow-geomechanics simulations are performed for these geomodels using GEOS, and these simulation results provide training data for the VAE. A VAE with stacked convolutional long short-term memory (ConvLSTM) layers is trained to parameterize pressure, strain, effective normal stress and shear stress fields at a set of time steps. The VAE parameterization is integrated with an ensemble smoother with multiple data assimilation (ESMDA) within the DSI framework to provide posterior predictions. Four in-situ monitoring wells provide the pressure and strain data used for history matching. The locations of these wells are determined through a separate optimization procedure. Posterior (history matched) results for three synthetic true models are presented. The results show that the DSI-VAE framework can provide accurate and efficient predictions for pressure, strain, and stress fields as well as for fault slip tendency. In addition, the framework is shown to reduce uncertainty in key parameters, such as Young's modulus, Poisson ratio, Biot coefficient, and fault permeability multipliers.