Rapid modelling of reactive transport in porous media using machine learning: limitations and solutions

Vinicius L S Silva,Geraldine Regnier,Pablo Salinas,Claire E Heaney,Matthew D Jackson,Christopher C Pain

2024-05-23

Abstract:Reactive transport in porous media plays a pivotal role in subsurface reservoir processes, influencing fluid properties and geochemical characteristics. However, coupling fluid flow and transport with geochemical reactions is computationally intensive, requiring geochemical calculations at each grid cell and each time step within a discretized simulation domain. Although recent advancements have integrated machine learning techniques as surrogates for geochemical simulations, ensuring computational efficiency and accuracy remains a challenge. This chapter investigates machine learning models as replacements for a geochemical module in a reactive transport in porous media simulation. We test this approach on a well-documented cation exchange problem. While the surrogate models excel in isolated predictions, they fall short in rollout predictions over successive time steps. By introducing modifications, including physics-based constraints and tailored dataset generation strategies, we show that machine learning surrogates can achieve accurate rollout predictions. Our findings emphasize that, when judiciously designed, machine learning surrogates can substantially expedite the cation exchange problem without compromising accuracy, offering significant potential for a range of reactive transport applications.

Computational Engineering, Finance, and Science

What problem does this paper attempt to address?

This paper discusses the use of machine learning to simulate reactive transport in porous media more quickly. Currently, calculating coupled fluid flow and chemical reactions is computationally expensive, requiring chemical reaction calculations at each grid cell and time step. Although previous studies have used machine learning techniques as alternative methods for chemical simulations, ensuring both efficiency and accuracy remains a challenge.

The research found that when predicting individually, machine learning agents perform well, but they encounter issues in predicting continuously over time steps (rollouts). To address this problem, the paper proposes incorporating physical constraints and customized dataset generation strategies to enhance the accuracy of machine learning agents in continuous time step predictions. Through this approach, machine learning agents can expedite cation exchange problems without sacrificing accuracy, offering potential for various reactive transport applications.

The paper first introduces geochemical simulators, fluid and transport simulators, and their coupling for simulation. Then, it studies different machine learning and deep learning strategies, as well as the impact of dataset sampling methods on prediction errors, and explores how to utilize physical knowledge of the problem to improve prediction accuracy. The experiments focus on simulating cation exchange reactions, a widely documented case, to evaluate the applicability of the models.

Through experiments, the authors found that machine learning agents perform well in one-shot predictions but fail due to accumulated errors (rollouts) during the simulation process. By using improved dataset generation methods and physical constraints, they demonstrate that machine learning agents can effectively replace geochemical modules, thus accelerating the simulation of cation exchange problems.

In conclusion, this paper aims to address the challenge of using machine learning to accelerate reactive transport simulations in porous media. It emphasizes the importance of considering physical constraints and appropriate dataset generation when designing agent models to ensure accuracy in long-term predictions.