Analysis of structural integrity of geological reservoirs for CO2 storage regarding to fault reactivation: a forecasting model based on machine learning techniques

Abstract The reactivation of geological faults can be triggered by stress changes due to fluid injection affecting their permeability and potentially transforming them into preferential flow channels. This can impact reservoir production dynamics and compromise fluid storage safety. This study assesses the feasibility of long-term CO2 storage in a geological reservoir in the Campos Basin, Rio de Janeiro, Brazil, focusing on fault system structural integrity vital in preventing leakage and ensuring storage security. This work develops a surrogate model for decision rules based on the Mohr–Coulomb criterion. Using geomechanical ranges derived from laboratory tests, a synthetic dataset was generated to train and simulate fault reactivation scenarios via Machine Learning (ML). This approach transforms geomechanical stability principles into an efficient decision framework. To ensure robustness, the predictive performance and sensitivity to training set size were evaluated by comparing Linear Discriminant Analysis (LDA), Random Forests (RF), and Artificial Neural Networks (ANN). The results obtained indicate an excellent predictive performance of the ML algorithms, obtaining better results with larger datasets, and presenting model accuracy of 97%, 98% and 99% for LDA, RF and ANN, respectively. The study proposes a workflow for developing robust predictive ML models to assess the integrity of reservoirs intersected by geological faults with respect to fluid storage. It acknowledges the limitations, and recommends simulations in geological formations for validation in different scenarios. In addition, the computational efficiency of this method surpasses conventional numerical simulation and allows generalization to other contexts, integrating specific characteristics of different scenarios.