Learned coupled inversion for carbon sequestration monitoring and forecasting with Fourier neural operators

Abstract

Seismic monitoring of carbon storage sequestration is a challenging problem involving both fluid-flow physics and wave physics. Additionally, monitoring usually requires the solvers for these physics to be coupled and differentiable to effectively invert for the subsurface properties of interest. To drastically reduce the computational cost, we introduce a learned coupled inversion framework based on the wave modeling operator, rock property conversion and a proxy fluid-flow simulator. We show that we can accurately use a Fourier neural operator as a proxy for the fluid-flow simulator for a fraction of the computational cost. We demonstrate the efficacy of our proposed method by means of a synthetic experiment. Finally, our framework is extended to carbon sequestration forecasting, where we effectively use the surrogate Fourier neural operator to forecast the CO 2 plume in the future at near-zero additional cost.

Real-time high-resolution CO2 geological storage prediction using nested Fourier neural operators

“…Once CO 2 is injected into the subsurface, machine learning models can be used for seismic inversion to provide gradients with respect to the inputs. 66 Due to the computational costs of high fidelity numerical simulation, inversion for existing CO 2 storage sites often rely on reduced models such as vertical equilibrium simulators. 67,68 Having rigorous analyses for these taskscan help reduce uncertainties and accelerate the CCS deployment scale-up progress.…”

Accelerating Carbon Capture and Storage Modeling using Fourier Neural Operators

“…We present the Nested FNO model for high-resolution 4D gas saturation and pressure buildup in CO 2 storage problems. The trained model provides exceptionally fast predictions and can support many engineering tasks that require repetitive forward simulations, including but not limited to (1) probabilistic assessment -as demonstrated above, (2) site selection 15 -quick screening for a large number of potential reservoirs, (3) storage optimizations 18,36 -exhaustive search in the parameter space, and (4) seismic inversion 37 -provide forward simulation outputs and gradients. Nested FNO can readily serve as a high-fidelity and high-resolution numerical simulator alternative and facilitate rigorous analyses for these tasks.…”

Model-Parallel Fourier Neural Operators as Learned Surrogates for Large-Scale Parametric PDEs

“…In addition to very fast simulation times during inference, FNOs and other deep learning-based approaches offer the possibility to compute gradients/sensitivities of PDEs using automatic differentiation (AD), thus making it possible to solve inverse problems without requiring users to manually differentiate and implement adjoints and/or gradients. Here, we highlight this opportunity with a recent example from [17] on subsurface CO 2 flow and seismic imaging. The goal of the example is to estimate subsurface medium parameters such as permeability from seismic data, which is an example of a coupled multi-physics problem.…”

“…Here we will forgo the rederivation of basic memory operations such as copy, clear, send, and receive as seen in [26], and instead focus on Figure 1: Coupled multi-physics inversion to estimate the subsurface permeability of a porous medium from seismic data measurements. To invert for the permeability, the authors in [17] first train a FNO that maps a permeability field to the CO 2 concentration history, which in turn is converted to the acoustic wave speed and used for simulating the seismic response. In the inverse problem, changes in the seismic data are first mapped to changes in the wave speed and the corresponding perturbations of the CO 2 concentration.…”