Prediction of long-term deformation in coal mine roadways for compressed air energy storage using numerical simulation and XGBoost

Compressed air energy storage (CAES) in abandoned coal mine roadways is a promising large-scale technology for supporting renewable energy integration. However, assessing the long-term stability of surrounding rocks under cyclic pressure loading remains challenging because conventional numerical simulations are computationally expensive and can only cover limited operating conditions. In this study, a FLAC3D model incorporating the Burgers creep model was developed using geological data from an actual coal mine. Based on an L 25 (5 5 ) orthogonal design, 25 operating scenarios were simulated over a 30-year period, considering burial depth, pressure limits, and operating frequency as key variables. An XGBoost regression model was then trained to predict long-term deformation of the roadway roof, sidewalls, and floor. The results showed that deformation and plastic failure increased with greater burial depth and longer low-pressure duration, whereas a higher operating frequency reduced cumulative deformation under the studied conditions. The XGBoost model achieved excellent predictive accuracy, with the coefficient of determination (R 2 ) exceeding 0.99 for roof subsidence prediction. SHAP analysis identified the lower-pressure limit and burial depth as the two most influential factors controlling roadway stability. The proposed data-driven framework provides an efficient approach for rapid stability evaluation and operational optimization of CAES systems in abandoned coal mine roadways.