Deep Learning Accelerated Well Control Optimization in Geologic CO2 Storage
Abstract
In geologic CO2 storage operations, the design of optimal CO2 injection well controls including injection rate/pressure and injection duration for each well is not trivial as it involves the consideration of practical constraints such as pressure buildup and CO2 plume migration. Thus, multiple objective functions are needed during the optimization process. The optimization problem is also challenged by the high computational cost associated with the full-physics numerical simulations. We developed a deep learning-based proxy model to improve the overall efficiency of well control optimization process in geologic CO2 storage. To consider practical CO2 storage operational constraints, this work focuses on maximization of CO2 storage and minimization of CO2 plume Area of Review (AoR). We leveraged Fourier Neural Operator (FNO) based deep learning algorithm to develop surrogate models to replace time-consuming full-physics flow simulations for the evaluations of bi-objective functions. The developed proxy models are integrated into the StoSAG algorithm, which is an efficient algorithm to solve field development optimization problem under geologic uncertainty. We will present application of the proposed workflow to several complex CO2 storage case studies to 1) evaluate the performance of the deep learning-based proxy model in terms of prediction accuracy and calculation efficiency and 2) to examine the final obtained control parameters for the maximization of CO2 injection as well as the minimization of plume AoR.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFMGC12E0487M