A Robust Deep Learning Workflow to Predict Multiphase Flow Behavior during Geological CO2 Sequestration Injection and Post-Injection Periods (2107.07274v1)

Abstract: This paper contributes to the development and evaluation of a deep learning workflow that accurately and efficiently predicts the temporal-spatial evolution of pressure and CO2 plumes during injection and post-injection periods of geologic CO2 sequestration (GCS) operations. Based on a Fourier Neuron Operator, the deep learning workflow takes input variables or features including rock properties, well operational controls and time steps, and predicts the state variables of pressure and CO2 saturation. To further improve the predictive fidelity, separate deep learning models are trained for CO2 injection and post-injection periods due the difference in primary driving force of fluid flow and transport during these two phases. We also explore different combinations of features to predict the state variables. We use a realistic example of CO2 injection and storage in a 3D heterogeneous saline aquifer, and apply the deep learning workflow that is trained from physics-based simulation data and emulate the physics process. Through this numerical experiment, we demonstrate that using two separate deep learning models to distinguish post-injection from injection period generates the most accurate prediction of pressure, and a single deep learning model of the whole GCS process including the cumulative injection volume of CO2 as a deep learning feature, leads to the most accurate prediction of CO2 saturation. For the post-injection period, it is key to use cumulative CO2 injection volume to inform the deep learning models about the total carbon storage when predicting either pressure or saturation. The deep learning workflow not only provides high predictive fidelity across temporal and spatial scales, but also offers a speedup of 250 times compared to full physics reservoir simulation, and thus will be a significant predictive tool for engineers to manage the long term process of GCS.

• The paper introduces a deep learning workflow using FNO to overcome high computational costs in simulating multiphase flow during CO2 sequestration.
• It splits the injection and post-injection periods to optimize prediction fidelity using cumulative CO2 injection volume as a key feature.
• The approach achieves up to 250x speedup with low RMSE, offering practical benefits for real-time subsurface management and inverse modeling.

Predicting Multiphase Flow Behavior During Geological CO2 Sequestration: A Deep Learning Approach

The paper titled "A Robust Deep Learning Workflow to Predict Multiphase Flow Behavior during Geological CO2 Sequestration Injection and Post-Injection Periods" addresses the challenges associated with simulating multiphase flow in porous media during geologic CO2 sequestration (GCS) processes. Traditional physics-based approaches, while generally effective, suffer from significant computational costs due to the inherent nonlinearity of the coupled physics involved. This paper introduces a deep learning (DL) workflow aimed at providing cost-effective and accurate predictions of pressure and CO2 saturation plumes during GCS operations.

Methodological Insights

At the core of the proposed approach is the Fourier Neural Operator (FNO), which allows for efficient handling of spatial topology predictions. The paper explores distinct DL models for CO2 injection and post-injection periods, acknowledging the different primary driving forces, such as viscous forces during injection and gravitational forces during post-injection relaxation. The researchers thoroughly evaluate different combinations of input features to optimize prediction fidelity for each period. Of significant importance is the cumulative CO2 injection volume, used as a feature to efficiently predict both pressure and saturation states.

The DL workflow combines several input variables, such as rock properties, operational controls, and time steps, to generate predictions in a 3D heterogeneous saline aquifer scenario. The paper leverages image-based approaches and utilizes FNO architecture that integrates both Fourier layers and fully connected layers to attain high-dimensional feature representations and sequential feature map updates.

Results and Performance Metrics

This research demonstrates the DL workflow's capability to offer a computational speedup by 250 times compared to full physics-based reservoir simulation methods without sacrificing prediction accuracy. For pressure predictions, the combined use of separate models during injection and post-injection periods yielded the lowest root mean square error (RMSE) values, signifying higher precision in capturing the dynamics of fluid movement.

The evaluation process revealed that during the post-injection period, the use of cumulative CO2 injection volume as a feature provided critical information for maintaining prediction precision, achieving an RMSE of 3.78 psi for pressure prediction. Similarly, saturation predictions utilizing the cumulative CO2 injection volume achieved an RMSE of 0.0254, indicating noteworthy accuracy across varying temporal and spatial scales.

Implications and Future Directions

The insights derived from this paper are poised to substantially influence the efficiency of geological CO2 sequestration modeling. Practically, the enhanced speed and accuracy of predictions present substantial benefits for real-time subsurface management. Furthermore, the reduced computational burden implies more feasibility for inverse modeling tasks that require vast amounts of forward simulation runs for uncertainty quantification and risk assessment.

Looking forward, continued advancements in DL frameworks such as FNO provide a promising avenue for enhanced modeling of intricate geophysical processes. Additionally, integration of broader geophysical datasets and adaptive learning paradigms could further refine model sensitivity to varying geological configurations and improve long-term management strategies for CO2 storage sites.

Ultimately, the outcomes of this paper underscore the compelling role of deep learning approaches in elevating traditional reservoir management methodologies, thus fostering more efficient and reliable GCS processes.