Evaluation Metrics for The Atlas CCS Project Using An AI Assisted Workflow
CCS project development requires a critical understanding of CO2 plume history, pressure footprints, and injectivity potential prior to execution. This study presents a creative approach to characterize subsurface properties dynamically and to quantify project uncertainties. We developed an AI assisted workflow to analyze CO2 plume sizes, pressure footprint, and injectivity potential quantitatively. We trained random forest proxy models to save the computational costs of massive Monte Carlo simulation and enable fast estimations of CO2 injection potential.
To understand the sensitivity of the identified parameters to design targets, we conducted massive full-physics reservoir simulation using the pyCCUS platform to perform 300 parallel computations on HPC to reduce the overall evaluation time. The parameters are selected according to prior experience, including porosity, permeability, relative permeability based on rock types, reservoir thickness, and injection schemes. The design targets are analyzed according to business needs, including CO2 plume size, pressure footprints, and the injectivity potential.
Results show the significance of each of the identified parameters on all variable responses. The random forest model was trained to predict the response variables in a fast manner. The proxy model shows 98% accuracy over the three variables to predict the injection potential. This workflow enables the Atlas CCS project planners to develop a full field injection scheme and define the management measurement and verification (MMV) program with improved confidence. Moreover, this workflow is capable to scale and be applied to other CCS projects to support decision making.