InterPore2026

Physics-Informed Modeling of Flow Instabilities during CO₂ Migration in Saline Aquifers

22 May 2026, 14:30

15m

Oral Presentation (MS03) Flow, transport and mechanics in fractured porous media MS03

Speaker

Henglai Zhai

Description

Geological carbon sequestration has been widely recognized as a promising strategy for mitigating CO₂ emissions by storing carbon dioxide in subsurface geological formations, such as saline aquifers. While recent studies have largely focused on optimizing CO₂ trapping mechanisms to improve storage efficiency, the flow dynamics of CO₂ plume migration—particularly the development of viscous and density-driven fingering instabilities arising from the strong contrasts between CO₂ and brine—remain insufficiently understood. Moreover, despite significant advances in computational resources and smart field technologies, the implementation of model-based operational and control strategies is still limited by the high computational cost, complexity, and limited accessibility of conventional high-fidelity simulations. Developing efficient and affordable modeling approaches is therefore essential, not only for improving the physical understanding of CO₂ plume dynamics in porous media, but also for enabling cost-effective deployment of control and decision-making strategies in geological carbon storage[1].

In this study, we propose a deep learning–based framework employing physics-informed neural networks (PINNs) to simulate flow dynamic instabilities in immiscible compressible multiphase flow systems relevant to CO₂ sequestration. The central idea of the PINN approach is to embed the governing physical laws, expressed as nonlinear partial differential equations (PDEs), directly into the neural network training process as physics-based constraints[2]. The governing equations are formulated based on conservation principles, and a deep neural network is constructed to approximate the solution fields, with spatial–temporal coordinates as inputs and the relevant flow variables as outputs. Using automatic differentiation, the PDE residuals are evaluated and incorporated into a composite loss function together with the initial and boundary conditions, enabling the network to satisfy the underlying physics without relying on traditional mesh-based discretization.

The proposed PINN framework is applied to solve the strongly nonlinear PDE system associated with flow instabilities during CO₂ plume migration. Its accuracy and robustness are assessed through systematic comparisons with high-fidelity numerical solutions obtained using the ICFERST framework, which is based on a control volume finite element method (CVFEM)[3]. The results demonstrate that PINNs can effectively capture key features of multiphase flow dynamics and instability evolution, highlighting their potential as a computationally efficient alternative for modeling and control-oriented simulations in geological carbon sequestration.