Machine learning–enabled optimization of a direct air capture system integrated with enhanced oil recovery

[Geoengineering News]

https://www.sciencedirect.com/science/article/pii/S2211715625008203

Authors: Farzin Hosseinifard, Shahabeddin Ghasemzadeh, Mohsen Salimi, Majid Amidpour

06 November 2025

https://doi.org/10.1016/j.rechem.2025.102836

Highlights

•Ensemble models achieved highest accuracy in EOR factor prediction.

•SHAP plots revealed top features driving model predictions.

•Residual analysis confirmed Random Forest's low bias and variance.

•DAC integration increased EOR factor from 19 % to 21.3 %.

•Study supports AI-based optimization for carbon-EOR strategies.

Abstract

The escalating levels of CO₂ in the atmosphere have heightened global environmental concerns, necessitating the deployment of efficient and scalable carbon extraction strategies. Among emerging methods, direct air capture (DAC) stands out as a viable approach. This research introduces a novel DAC configuration tailored to enhance the efficiency of enhanced oil recovery (EOR). The DAC system was modeled using Aspen Plus V11, employing a hydroxide-to‑carbonate conversion pathway for CO₂ absorption. As part of broader carbon management efforts, Carbon Capture, Utilization, and Storage (CCUS) plays a pivotal role in curbing emissions, particularly through its application in subsurface oil recovery processes.

To assess and forecast the impact of DAC-sourced CO₂ on EOR performance in Abadan, a suite of Machine learning techniques was applied. These included XGBoost, Random Forest, Gradient Boosting, Support Vector Regression, Linear Regression, k-Nearest Neighbors, Bagging, and Stacking. Among the models tested, the Decision Tree algorithm demonstrated the highest predictive capability, yielding an R2 score of 0.87. It effectively estimated a growth in EOR efficiency from 19 % to approximately 21.3 %.

Source: ScienceDirect