InterPore2026

A High-Fidelity Surrogate for Multiphase Flow in Complex Faulted System Using Geometric-Aware Fourier Neural Operator

19 May 2026, 15:05

1h 30m

Poster Presentation (MS15) Machine Learning in Porous Media Poster

Speaker

Ching-En Kung (National Taiwan University)

Description

As a viable solution to climate change, carbon capture and storage (CCS) plays a crucial role in achieving net-zero emissions. Injecting CO2 into deep geological formations leads to fluid pressure buildup and CO2 plume migration, which may induce seismic events or contaminate groundwater resources. These hazards necessitate risk assessment and storage prospect evaluations, which rely heavily on forecasts of subsurface flow processes.

However, traditional numerical approaches could be computationally prohibitive, especially when performing uncertainty analysis for complex, heterogeneous subsurface environments. While Fourier Neural Operator (FNO) has emerged as a high-speed surrogate model, its reliance on Fast Fourier Transform restricts its applications to structured grids. This poses a significant limitation for geological models where unstructured grids are necessary to characterize complex fault and fracture structures. Such limitations are particularly relevant for tectonically active storage site.

Taoyuan, Taiwan has been considered as a potential site for underground geological storage due to its thick sedimentary rock formations overlain by a shale caprock. Nevertheless, owing to locate at the convergent boundary between the Eurasian plate and the Philippine Sea plate makes it one of the most seismically active regions in the world. Seismic profiles from the area also indicate the presence of several possible faults, which may pose challenges for long-term storage security.

In this work, we propose Geometric-aware Fourier Neural Operator to efficiently evaluate the storage potential and relevant risk at the Taoyuan site. The basic geological model is supported by core data and seismic profiles. To build a robust training and validation dataset, we deploy well-known multiphase flow simulator TOUGH2 with ECO2N module, simulating CO2 injection across varying depth in 50 years. Also, the dataset encompasses a wide range of stochastic geophysical properties distributions and diverse fault architecture to account for subsurface uncertainty. The expected results provide a high-accuracy surrogate for multiphase fluid simulation in complex fault systems, enabling sensitivity studies to be several orders of magnitude faster than the traditional solvers.