Survival Analysis-Powered AI in Risk-Aware Data-Driven Multi-Objective Optimization for Geological CO2 Storage
SPE Latin American and Caribbean Petroleum Engineering Conference (LACPEC), 2026
We combine survival analysis-based surrogate models with multi-objective optimization to identify CO2 injection strategies that balance storage volume against long-term leakage risk. The framework uses the NSGA-II genetic algorithm and computationally inexpensive surrogates to explore competing objectives under geological uncertainty.
Survival analysis accounts for censored observations from shortened reservoir simulations, reducing the simulation effort required to train the surrogate. We evaluate the approach in a synthetic saline aquifer case study with four CO2 injectors and five pressure-maintenance producers, demonstrating its potential to support efficient, risk-informed storage and reservoir management.
Recommended citation: Gurwicz, A., Alves Abreu, A. C., Gutman, D. H., Gildin, E., & Cavalcanti Pacheco, M. A. (2026). Survival analysis-powered AI in risk-aware data-driven multi-objective optimization for geological CO2 storage. SPE Latin American and Caribbean Petroleum Engineering Conference (LACPEC), Rio de Janeiro, Brazil, June 9–11. Paper SPE-231654-MS. https://doi.org/10.2118/231654-MS