[Paper Review] A Digital Twin for Geological Carbon Storage with Controlled Injectivity

Abhinav Prakash Gahlot, Haoyun Li+3 more|arXiv (Cornell University)|Mar 28, 2024

Scientific Computing and Data Management4 citations

TL;DR

This paper presents a digital twin for geological carbon storage that integrates uncertainty-aware generative AI with real-time monitoring data to control CO₂ injectivity and prevent cap rock fracturing. By using sequential Bayesian inference and neural posterior estimation on time-lapse seismic and well data, the system optimizes injection rates to keep fracture probability below 1% with 97.5% confidence, reducing risk from 32.59% to 1% when injection is lowered to 0.0387 m³/s.

ABSTRACT

We present an uncertainty-aware Digital Twin (DT) for geologic carbon storage (GCS), capable of handling multimodal time-lapse data and controlling CO2 injectivity to mitigate reservoir fracturing risks. In GCS, DT represents virtual replicas of subsurface systems that incorporate real-time data and advanced generative Artificial Intelligence (genAI) techniques, including neural posterior density estimation via simulation-based inference and sequential Bayesian inference. These methods enable the effective monitoring and control of CO2 storage projects, addressing challenges such as subsurface complexity, operational optimization, and risk mitigation. By integrating diverse monitoring data, e.g., geophysical well observations and imaged seismic, DT can bridge the gaps between seemingly distinct fields like geophysics and reservoir engineering. In addition, the recent advancements in genAI also facilitate DT with principled uncertainty quantification. Through recursive training and inference, DT utilizes simulated current state samples, e.g., CO2 saturation, paired with corresponding geophysical field observations to train its neural networks and enable posterior sampling upon receiving new field data. However, it lacks decision-making and control capabilities, which is necessary for full DT functionality. This study aims to demonstrate how DT can inform decision-making processes to prevent risks such as cap rock fracturing during CO2 storage operations.

Motivation & Objective

• To develop a digital twin for geologic carbon storage that enables real-time monitoring and uncertainty quantification using multimodal data.
• To address the risk of cap rock fracturing during CO₂ injection by integrating decision-making into the digital twin framework.
• To optimize injection rates based on probabilistic constraints, ensuring fracture probability remains below a predefined threshold.
• To bridge siloed domains like geophysics and reservoir engineering through a unified, data-driven virtual replica.
• To demonstrate how generative AI and Bayesian inference can enable proactive risk mitigation in carbon storage operations.

Proposed method

• The digital twin uses recursive training with simulated state-observation pairs derived from posterior sampling at each timestep.
• Neural posterior density estimation via simulation-based inference enables uncertainty-aware state estimation from time-lapse seismic and well data.
• Sequential Bayesian inference is applied to update the posterior distribution of reservoir states and permeability at each time step.
• A constrained optimization problem is solved per sample to maximize injectivity rate while ensuring reservoir pressure does not exceed depth-dependent fracture pressure.
• Kernel density estimation (KDE) smooths the empirical fracture frequency distribution to derive a cumulative distribution function (CDF) for decision support.
• Confidence intervals for fracture probability are computed using the Bernoulli distribution and a 95% confidence level (Z = 1.96), with conservative (left-tailed) selection of injection rates.

Experimental results

Research questions

• RQ1 How can a digital twin integrate multimodal monitoring data (seismic and well observations) to enable uncertainty-aware state estimation in geological carbon storage?
• RQ2 What injection rate minimizes the risk of cap rock fracturing while maintaining high CO₂ injectivity?
• RQ3 How can Bayesian inference and generative AI be used to quantify uncertainty in reservoir pressure and permeability for risk-informed decision-making?
• RQ4 Can a digital twin provide confidence intervals for fracture probability to support conservative operational decisions?
• RQ5 How does controlling injectivity using uncertainty-aware inference reduce the number of fractured samples compared to uncontrolled injection?

Key findings

• The initial injection rate of 0.0500 m³/s results in a fracture probability of 32.59% with a 95% confidence interval of 24.47% to 40.71%, indicating unacceptably high risk.
• Lowering the injection rate to 0.0387 m³/s reduces the fracture probability to less than 1% while maintaining 97.5% confidence in the result.
• Out of 128 state samples, 43 were fractured at the initial injection rate, compared to only one fractured sample at the controlled rate.
• The digital twin successfully identifies a conservative injection rate that limits fracture occurrence to 1% with high statistical confidence.
• The use of KDE and CDF analysis enables precise, data-driven selection of injection rates that balance injectivity and safety.
• The system demonstrates that uncertainty-aware digital twins can transition from passive monitoring to active risk mitigation through controlled injectivity.