Simulating CO2 storage under geomechanical risks frequently involves substantial computational costs due to the coupling between multiphase flow and geomechanics. Implementing standard workflows, such as well location optimization, with such coupled physics models can significantly increase the computational overhead and make the models impractical to use. We study the feasibility of using deep-learning models to significantly reduce the computational overhead associated with simulating and quantifying the geomechanical risks of CO2 storage. The proposed approach leverages deep learning-based surrogate modeling to significantly enhance the efficiency of coupled flow-geomechanics simulations for identifying suitable injection well locations for CO2 storage. Using simulated data, we train a U-Net convolutional neural network to learn a mapping between well locations s and spatially distributed model parameters (permeability) to the simulation outputs of interest. Once trained with a fixed set of model input parameters, the U-Net model can map different well location scenarios to the corresponding pressure fields, CO2 saturation, and geomechanical outputs, including vertical displacement and plastic strain. The U-Net model is subsequently adopted as an efficient tool to replace the coupled flow-geomechanics simulation needed for identification of injection well locations to minimize geomechanical risks. We report preliminary results showing that the trained U-Net model can predict pressure and saturation fields from well locations, with all the other inputs remaining consistent with the simulation model used in the training. We investigate the performance of the network under different assumptions and for estimating different flow and geomechanical outputs. The results show that the U-Net model can drastically reduce the computational cost of well placement workflows by replacing coupled physics simulation with a fast proxy model that can be used to predict the geomechanical risk associated with different well location and injection strategies. The developed framework can be used to improve the computational demand of coupled-physics modeling and facilitate its application to decision-making workflows and field management.
Skip Nav Destination
SPE/AAPG/SEG Carbon, Capture, Utilization, and Storage Conference and Exhibition
March 11–13, 2024
Houston, Texas, USA
ISBN:
978-1-959025-62-7
A Deep Learning-Based Surrogate Model for Rapid Assessment of Geomechanical Risks in Geologic CO2 Storage Available to Purchase
Fangning Zheng;
Fangning Zheng
University of Southern California
Search for other works by this author on:
Birendra Jha;
Birendra Jha
University of Southern California
Search for other works by this author on:
Behnam Jafarpour
Behnam Jafarpour
University of Southern California
Search for other works by this author on:
Paper presented at the SPE/AAPG/SEG Carbon, Capture, Utilization, and Storage Conference and Exhibition, Houston, Texas, USA, March 2024.
Paper Number:
SPE-CCUS-2024-4003166
Published:
March 11 2024
Citation
Zheng, Fangning, Jha, Birendra, and Behnam Jafarpour. "A Deep Learning-Based Surrogate Model for Rapid Assessment of Geomechanical Risks in Geologic CO2 Storage." Paper presented at the SPE/AAPG/SEG Carbon, Capture, Utilization, and Storage Conference and Exhibition, Houston, Texas, USA, March 2024. doi: https://doi.org/10.15530/ccus-2024-4003166
Download citation file:
Sign in
Don't already have an account? Register
Personal Account
You could not be signed in. Please check your username and password and try again.
Could not validate captcha. Please try again.
Pay-Per-View Access
$28.00
Advertisement
103
Views
Advertisement
Suggested Reading
Advertisement