Deep Learning Assisted Multi-Objective Optimization of Geological CO2 Storage Under Geomechanical Risks

Abstract In geological CO2 storage, designing the optimal well control strategy for CO2 injection to maximize CO2 storage while minimizing the associated geomechanical risks is not trivial. This challenge arises due to pressure buildup, CO2 plume migration, the highly nonlinear nature of geomechanical responses during CO2 injection on rock-fluid interaction, as well as the high computational cost associated with coupled flow-geomechanics simulations. In this paper, we introduce a novel optimization framework to address these challenges. The optimization problem is formulated as follows: maximize total CO2 storage while minimizing geomechanical risks by adjusting the injection well's bottom-hole pressure (BHP) within bounded constraints. The geomechanical risks are primarily driven by injection-induced pressure build-up, which is characterized by the seismic moment. We employ the Fourier Neural Operator (FNO) based deep learning model to construct surrogate models, replacing the time-consuming coupled flow-geomechanics simulations for evaluating the objective function. The input features of the FNO-based proxy model comprise flow properties and well controls. Meanwhile, the outputs consist of time-series reservoir pressure and CO2 saturation as the state variables. The FNO-based proxy model is trained on a synthetic case simulating large-scale CO2 storage. The results demonstrate that the deep learning-based proxy models accurately and efficiently predict the designated state variables. The developed proxy models have been incorporated into a multi-objective optimization framework through a Genetic Algorithm (NSGA-II) to reduce the computational burden. The proposed optimization framework exhibits significant computational acceleration compared to conventional simulation-based function evaluation strategies. A set of Pareto-optimal solutions of the proposed workflow yields nontrivial optimal decisions, diverging from scenarios where geomechanical risks associated with CO2 injection are disregarded. Our findings reveal that the injection rates may not need to reach their upper bounds to achieve the highest storage capacity when geomechanical risks are considered. This is significant because pushing injection rates too high can induce rock failure and unacceptable levels of micro-seismic and leakage events. Thus, the maximum storage capacity is constrained to maintain the reservoir in an elastic state. The novelty of this work is twofold. First, we demonstrate the importance of incorporating the minimization of the geomechanical risks as an objective function into the CO2 storage optimization workflow to mitigate the potential risk of induced micro-seismicity and CO2 leakage. Second, leveraging the FNO-based proxy model to optimize a real CO2 storage operation represents a new approach.