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

TL;DR

This paper explores using machine learning as a surrogate for geochemical calculations in reactive transport simulations, identifying limitations in predictive rollouts and proposing physics-based modifications to improve accuracy.

Contribution

It demonstrates the limitations of ML surrogates in reactive transport modeling and introduces physics-based constraints to enhance their predictive capabilities over multiple time steps.

Findings

ML surrogates excel in isolated predictions but struggle with rollout predictions.

Physics-based modifications improve the accuracy of ML surrogates in successive time-step predictions.

Even simple geochemical reactions pose challenges for ML models without tailored constraints.

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 work investigates machine learning models as replacements for a geochemical module in a simulation of reactive transport in porous media. As a proof of concept, 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 even for a simple sorption equilibrium reaction (cation exchange problem), machine learning surrogates alone fail in predicting over successive time-steps. Incorporating simple physics-based modifications enables us to overcome this limitation. A detailed analysis of the limitations and potential mitigation strategies is presented in this work.