Physics-Informed Spatiotemporal Surrogate Modeling and Robust Operational Optimization for CO₂ Applications in Subsurface Flow under Permeability Uncertainty

Overview

CO₂ applications in subsurface formations require accurate prediction, characterization, and operational control under geological uncertainty. High-fidelity numerical simulations can describe coupled multiphase flow, pressure propagation, plume migration, trapping, and thermal transport, but their computational cost limits repeated uncertainty analysis, data assimilation, and operational optimization.

This talk presents physics-informed deep-learning workflows that combine numerical simulation data with governing physical constraints. Unlike purely data-driven models, which may generate visually plausible field maps while creating or removing mass or violating energy conservation and pressure–flow relationships, the proposed framework incorporates CO₂ mass conservation, thermal-energy balance, and Darcy-flow consistency into model training. The resulting spatiotemporal surrogate models rapidly predict pressure, CO₂ saturation, residual trapping, and temperature fields across heterogeneous reservoirs and varying well-control conditions.

Once trained, these surrogates can generate physically consistent field predictions hundreds of times faster than conventional numerical simulations. This acceleration enables large numbers of forward evaluations, making iterative data assimilation for heterogeneous permeability inversion and ensemble-based optimization of staged injection and production strategies computationally practical within hours. Applications include geological CO₂ storage, CO₂ flooding for enhanced oil recovery, geothermal heat recovery, pressure management, and economic evaluation.