Advances in Machine-Learning-Driven CO₂ Geological Storage: A Comprehensive Review and Outlook

Abstract

As the cornerstone of carbon capture, utilization, and storage (CCUS) technologies, CO 2 geological storage (CGS) enables atmospheric carbon isolation through subsurface storage, representing a critical pathway toward achieving carbon neutrality. However, conventional physical models for CGS systems face persistent limitations, including computational inefficiency and predictive inaccuracies, when addressing reservoir heterogeneity, migration pathway uncertainties, and long-term security risks. Machine learning (ML) has emerged as a transformative solution to these problems. This review systematically synthesizes the progress in ML applications over the last 5 years for CGS through a four-dimensional framework: (1) algorithmic innovation, comparing classical ML and deep learning techniques tailored for CGS; (2) laboratory-scale breakthroughs, highlighting ML-driven parameter prediction and multimodal data fusion for experimental characterization; (3) simulation enhancements, showcasing ML-augmented geological inversion, trapping mechanism analysis, and multiphase flow modeling; and (4) field-scale implementations, including ML-optimized wellbore management, real-time leakage monitoring, and risk benefit assessments. Although the findings reveal substantial progress in ML-based CGS integration, they underscore enduring challenges in data quality, model interpretability, and scalability. By critically evaluating these advancements, identifying research gaps, and proposing future research directions, this work establishes a theoretical foundation for synergistic ML-based CGS integration. The insights presented in this Perspective aim to accelerate the deployment of scalable, intelligent CCUS systems, directly supporting global decarbonization efforts.

Topics & Concepts