<p>Underground caverns have emerged as a prominent choice for gas storage site selection, owing to their economic efficiency and potential for large-scale capacity. During the operation of a Compressed Air Energy Storage (CAES) facility, rapid fluctuations in high pressure within the underground cavern promote the initiation and propagation of cracks, which can potentially compromise the cavern’s ultimate load-bearing capacity and, in turn, jeopardize overall structural stability. To elucidate the fracture mechanisms of surrounding rock under complex stress regimes, this study establishes a failure analysis framework for tunnel-type underground caverns employed in CAES, utilizing the Discrete Element Method (DEM). Concurrently, a machine learning-driven DEM surrogate model is developed to facilitate rapid prediction of fracture surface morphology parameters and to conduct sensitivity analyses of the governing factors. The results indicate that the fracture surface morphology is predominantly governed by the in situ stress ratio (<i>k</i>). Under a given <i>k</i> condition, different parameters (cohesion <i>c</i>, internal friction angle <i>φ</i>, cavern depth <i>H</i>, and cavern diameter <i>D</i>) contribute differently to fracture evolution. For <i>k</i> = 0.5 and <i>k</i> = 3, the influence of rock parameters on fracture surface morphology is negligible, respectively. However, for <i>k</i> = 1 and<i> k</i> = 2, <i>c</i> and <i>φ</i> emerge as the dominant parameters influencing fracture surface morphology, These findings deliver critical technical insights for enhancing the storage capacity and safety performance of liners in CAES underground caverns, and offer a valuable reference for the design and implementation of next-generation new energy infrastructure projects.</p>