Data-Driven Framework for Fast Screening and Multiobjective Optimization of CO ₂ Storage: Learning Temporal Evolution from Global Case Studies

High-fidelity numerical simulations of CO2 geological storage are computationally expensive, impeding the rapid assessment of leakage risks associated with mobile free gas. To address this, this study proposes a comprehensive data-driven framework ranging from mechanism analysis to parameter optimization. Leveraging a data set of 4,631 time-series samples derived from representative global sequestration projects, we constructed a generalized predictive model incorporating nine key geological and operational parameters. Among the tested algorithms, XGBoost demonstrated superior performance in capturing the temporal evolution of residual (RTI) and solubility trapping indices (STI). Crucially, the physics-based SHAP analysis quantified the geological controls: RTI is primarily driven by residual gas saturation via capillary hysteresis, while STI is strictly governed by formation water salinity, consistent with the thermodynamic limits of Henry’s Law. Furthermore, coupling the surrogate model with the NSGA-II algorithm generated Pareto-optimal solutions that effectively balance short-term safety and long-term stability. The optimization results establish strict quantitative screening criteria: formations with high permeability (>400 mD), ultralow salinity (<5 kppm), and moderate residual gas saturation (<0.22) are identified as optimal for safely restricting the initial mobile gas index to below 30% within the first 50 years. Large-scale Monte Carlo simulations (100,000 evaluations) confirm the robustness of these criteria. This framework bridges data-driven insights with physical mechanisms, providing a cost-effective tool for prefeasibility analysis and optimized scenario generation in CCUS projects.