<p>Understanding the three dimensional (3D) in-situ stress field is essential for the stability of underground caverns during excavation. However, accurately determining the in-situ stress field through most existing inversion analysis methods presents significant difficulties, thereby complicating the precise simulations of excavation damage zone (EDZ) based on these inversions. In this study, a machine learning algorithm is employed to perform 3D in-situ stress field inversion. Additionally, the submodel method is utilized to simulate the EDZ distribution under realistic stress conditions. The methodology was validated at the underground powerhouse of the Wuhai Pumped Storage Power Station, and its reliability was demonstrated in accurately inverting the 3D in-situ stress field and simulating the EDZ distribution, including identifying potential spalling locations. This study provides a novel approach for calculating the EDZ based on realistic in-situ stress field conditions, offering valuable insights for improving the stability of underground excavations.</p>