Abstract: As a key branch of emerging energy storage technologies, underground compressed air energy storage (CAES) is gaining increasing attention for its advantages in large-scale capacity, long-duration storage, and environmental sustainability, making it a crucial support for new power systems. However, underground CAES projects often face challenges such as complex geological conditions, significant multi-physical field coupling, and frequent injection-production cycles, where traditional methods show clear limitations in modeling accuracy and operational efficiency. In recent years, artificial intelligence (AI) technologies, with their powerful nonlinear modeling and data-driven capabilities, have offered novel approaches to intelligent site selection, structural prediction, system operation, and risk warning in underground CAES. This paper employs bibliometric analysis and knowledge mapping techniques to systematically review the current state of AI applications in underground CAES, covering typical scenarios such as site selection and geological modeling, intelligent cavern construction, stability prediction, injection-production optimization, multiphysical coupling modeling, and safety monitoring. The findings reveal that research in this field remains in its early stages, lacking a comprehensive and systematic framework. Based on existing studies, this paper proposes several key directions for future development, including physics-informed modeling, multi-source data integration, and the construction of intelligent engineering platforms, aiming to provide theoretical insights and technical references for advancing the intelligent development of underground CAES and supporting the realization of China’s dual carbon goals.

Key words: compressed air energy storage, rock and soil mechanics, artificial intelligence, underground engineering, machine learning