Abstract:CO$_2$ sequestration is a crucial engineering solution for mitigating climate change. However, the uncertain nature of reservoir properties, necessitates rigorous monitoring of CO$_2$ plumes to prevent risks such as leakage, induced seismicity, or breaching licensed boundaries. To address this, project managers use borehole wells for direct CO$_2$ and pressure monitoring at specific locations. Given the high costs associated with drilling, it is crucial to strategically place a limited number of wells to ensure maximally effective monitoring within budgetary constraints. Our approach for selecting well locations integrates fluid-flow solvers for forecasting plume trajectories with generative neural networks for plume inference uncertainty. Our methodology is extensible to three-dimensional domains and is developed within a Bayesian framework for optimal experimental design, ensuring scalability and mathematical optimality. We use a realistic case study to verify these claims by demonstrating our method's application in a large scale domain and optimal performance as compared to baseline well placement.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: In carbon dioxide sequestration (CO₂ sequestration) projects, how to optimally arrange the positions of monitoring wells in order to maximize the monitoring effect within a limited budget, ensure effective monitoring of the dynamics of the CO₂ plume, and prevent risks such as leakage, induced earthquakes, or breaching of permitted boundaries. ### Problem Background Carbon dioxide sequestration is an important engineering solution to address climate change. However, due to the uncertainty of reservoir properties, strict monitoring of the CO₂ plume is required to prevent potential risks. Project managers usually use drilling to directly monitor CO₂ and pressure at specific locations. Given the high cost of drilling, it is necessary to strategically select the positions of a small number of monitoring wells to ensure the most effective monitoring within budget constraints. ### Core Method of the Paper To solve the above - mentioned problems, the authors propose a method named BEACON, namely "Bayesian Experimental design Acceleration with Conditional Normalizing flows". This method combines a fluid - flow solver and a generative neural network to predict plume trajectories and infer the uncertainty of the plume. Specifically: 1. **Expected Information Gain (EIG)**: - Use Bayesian principles to select well locations that can provide the maximum information gain. EIG is defined as the difference (measured by Kullback - Leibler divergence) between the prior distribution $ p(x_k) $ and the posterior distribution $ p(x_k|y_k) $: \[ \text{EIG}(M)=\mathbb{E}_{p(y_k|M)}[D_{\text{KL}}(p(x_k|y_k)\|p(x_k))] \] 2. **Conditional Normalizing Flows**: - Utilize the normalizing - flow model to directly measure the Kullback - Leibler divergence, thereby optimizing well - location selection. By jointly optimizing the normalizing - flow parameters $ \theta $ and the well - location probability density $ w $, EIG is maximized: \[ \hat{\theta},\hat{w}=\argmax_{\theta,w}\frac{1}{N}\sum_{i = 1}^N\left(-\frac{1}{2}\|f_\theta(x_k^{(i)};M(w)\odot y_k^{(i)})\|^2_2+\log|\det J_{f_\theta}|\right) \] 3. **Training Data Generation**: - Use a fluid - mechanics solver to generate training samples, including forecasts at time step $ k $ derived from prior samples $ x_{k - 1} $ and adding noise $ \epsilon $ to simulate observed values $ y_k $. 4. **Iterative Optimization**: - The BEACON algorithm is integrated with Digital Twin, and the well - location selection is gradually optimized by iteratively updating the prior and posterior distributions. ### Results and Conclusions Through synthetic case studies, the authors verified the effectiveness of the BEACON algorithm. The results show that, compared with randomly selected well locations, BEACON can not only reduce the uncertainty of plume inference but also significantly reduce the error rate. This indicates that BEACON has important value in optimizing the monitoring and management of CO₂ sequestration projects. ### Summary This paper aims to develop an efficient and scalable method by combining Bayesian experimental design and conditional normalizing flows to optimize the layout of monitoring wells in CO₂ sequestration projects, thereby improving the monitoring effect and reducing costs.

BEACON: Bayesian Experimental design Acceleration with Conditional Normalizing flows $-$ a case study in optimal monitor well placement for CO$_2$ sequestration

Bayesian Monitoring Design for Streambed Heat Tracing: Numerical Simulation and Sandbox Experiments.

Inference of CO2 flow patterns -- a feasibility study

Improving Parameter Estimation with an Efficient Sequential Probabilistic Collocation-Based Optimal Design Method

Seismic monitoring of CO2 plume dynamics using ensemble Kalman filtering

Probabilistic Bayesian optimal experimental design using conditional normalizing flows

An optimization approach to design monitoring schemes for CO2 leakage detection

Three‐Axis Borehole Gravity Monitoring For Co2 Storage Using Machine Learning Coupled To Fluid Flow Simulator

Graph Network Surrogate Model for Optimizing the Placement of Horizontal Injection Wells for CO2 Storage

A deep learning-accelerated data assimilation and forecasting workflow for commercial-scale geologic carbon storage

Normalizing Flows for Bayesian Posteriors: Reproducibility and Deployment

A multi-dimensional parametric study of variability in multi-phase flow dynamics during geologic CO2 sequestration accelerated with machine learning

Assimilation of Geophysics-Derived Spatial Data for Model Calibration in Geologic CO2 Sequestration

Optimization of Pressure Management Strategies for Geological CO2 Sequestration Using Surrogate Model-based Reinforcement Learning

Improving monitoring network design to detect leaks at hazardous facilities: Lessons from a CO2 storage site

A Bayesian Approach for In-Situ Stress Prediction and Uncertainty Quantification for Subsurface Engineering

Bayesian Adaptive Calibration and Optimal Design

A Feasibility Study of CO2geologic Sequestration Integrating Reservoir Simulation, Rock-Physics Theory, and Seismic Modeling

Reservoir parameter inversion of CO2 geological sequestration based on the self-adaptive hybrid genetic algorithm

Variational Bayesian experimental design for geophysical applications: seismic source location, amplitude versus offset inversion, and estimating CO2 saturations in a subsurface reservoir

Ensemble-Based Closed-Loop Optimization Applied to Brugge Field