Abstract

Pore-scale reactive transport modeling is a powerful tool used to analyze micro-scale processes where fluid flow and geochemical reactions occur. Despite its capability to examine complex hydrological and geochemical system behavior, the high computational demands for these simulations present a significant limitation. To overcome this challenge, this study evaluated artificial neural network (ANN)-based surrogate models to replace geochemical reaction calculations, which consume the majority of computational time in reactive transport simulations. The study considered two ANN models: a combined model (CM) that simultaneously accounts for mineral dissolution/precipitation and solute adsorption reactions, and an independent model (IM) that treats these reactions independently. The performance of these models was compared using metrics, including mean squared error (MSE), coefficient of determination (R²), and mass balance errors. Results indicate that IM demonstrates superior accuracy compared to CM. This finding suggests that instead of constructing a single complex model for the entire geochemical reaction network, pore-scale geochemical reactions can be effectively replaced by combining individual neural network models trained for specific reactions.

Keywords geochemical reactions, artificial neural network, surrogate model, reactive transport, pore scale