Optimizing Carbon Utilization and Sequestration Efficacies in Sandstone Oil Reservoirs Using an Integrated Machine-Learning Workflow

In this study, a robust optimization framework incorporated with multiple machine learning algorithms is structured to co-optimize a CO₂ enhanced oil recovery (EOR) project considering carbon sequestration and hydrocarbon extraction efficacies. A field-scale numerical simulation model is employed to simulate a water alternative CO2 injection (WAG) project in Pennsylvanian Upper Morrow sandstone reservoir in the Farnsworth Unit (FWU), Texas.

A compositional model that couples numerous field data (geological, geophysical and engineering data) is established to serve for the optimization scheme. The model was calibrated based upon the field data to generate a successful history-matched model. The overall objective function for the optimization study is assigned by combining parameters such as oil recovery factor, CO₂ storage and net present value. In this work, neural network models are trained and employed as proxies of high-fidelity numerical models to reduce the extensive computational overheads involved in the optimization process. Simulation data generated from high-fidelity model were utilized to train the proxies via iterative calibration runs. Once an accepted proxy was realized, were it would be applied in evolutionary and machine learning optimization algorithms to find the optimal solution for the pre-defined objective function. Engineering design parameters to be optimized include well bottomhole operational pressure, water and gas injection rate, WAG cycle durations, etc. The robust optimization workflow developed in this work is competent to find the field development strategies that maximize both hydrocarbon production and CO₂ sequestration in FWU. It takes advantage of the fast computational speed of the neural network models to assist numerical simulations for field-scale CO₂-EOR projects. Moreover, the observations and lessons earned in this work would provide significant insights to similar projects throughout the world.