Spatial–temporal prediction of minerals dissolution and precipitation using deep learning techniques: An implication to Geological Carbon Sequestration. (1st June 2023)

• Record Type: Journal Article
• Authors: Tariq, Zeeshan; Yildirim, Ertugrul Umut; Gudala, Manojkumar; Yan, Bicheng; Sun, Shuyu; Hoteit, Hussein
• Is Part Of: Fuel. Volume 341(2023)
• Publication Date: 2023-06-01
• Subjects: Deep learning -- Numerical simulation -- CO2 mineralization -- Reactive transportation -- Big data analysis; Fuel -- Periodicals; Coal -- Periodicals; Coal; Fuel; Periodicals; 662.6
• DOI: 10.1016/j.fuel.2023.127677
• Languages: English
• ISSNs: 0016-2361

Abstract

Abstract: In Geological Carbon Sequestration (GCS), mineralization is a secure carbon dioxide ( CO 2 ) trapping mechanism to prevent possible leakage at a later stage of the GCS project. Modeling the mineralization mechanism during GCS relies on numerical reservoir simulation, but the computational cost is prohibitively high due to the complex physical processes. Therefore, deep learning (DL) models can be used as a computationally cheaper and more reliable at the same time, alternative to conventional numerical simulations. In this work, we have developed a DL approach to effectively predict the dissolution and precipitation of various essential minerals, including Anorthite, Kaolinite, and Calcite, during CO 2 injection into deep saline aquifers. We have established a reservoir model to simulate the geological CO 2 storage process. Seven hundred twenty-two numerical realizations were performed to generate a comprehensive dataset for training DL models. Two convolution neural networks (CNN), Fourier Neural Operator (FNO), and U-Net were trained. The trained models used reservoir and well properties along with time information as input and predicted the precipitation and dissolution of minerals in space and time scales. During the training process, root-mean-squared-error (RMSE) was used as a loss function. To gauge prediction performance, we have applied the trained model to predict the concentrations of different minerals on the test dataset, which is 15% of the entire dataset, and two metrics, including the average absolute percentage error (AAPE) and the coefficient of determination ( R 2 ), were adopted. The FNO model resulted in the R 2 of 0.95 for the Calcite model, 0.94 for the Kaolinite model, and 0.93 for the Anorthite model. The U-Net model resulted in the R 2 of 0.88 for the Calcite model, 0.89 for the Kaolinite model, and 0.912 for the Anorthite model. The model's prediction CPU time (0.2 s/case) was much lower than that of the physics-based reservoir simulator (3600 s/case). Therefore, the proposed method offers predictions as accurate as our physics-based reservoir simulations while providing a substantial computational time acceleration. Graphical abstract: Highlights: A robust deep learning (DL) workflow is presented. DL workflow can efficiently predict the spatial and temporal mineralization process. DL workflow showed substantial acceleration compared to full numerical reservoir simulation. Multi model approach enhanced the prediction performance of CO2 mineralization process.