To ensure the construction safety of geotechnical engineering in deep high stress areas, a combined rock burst intensity prediction model based on whale optimization algorithm (WOA) and extreme gradient boosting (XGBoost) is proposed to address the suddenness and complexity of rock burst. Firstly, the main controlling factors that affect the intensity level of rock burst are analyzed, and the uniaxial compressive strength, maximum tangential stress, uniaxial tensile strength, brittleness coefficient, stress coefficient, and elastic energy index are selected to establish a prediction index system for rock burst intensity level. The original samples are processed using the Pearson correlation coefficient, multiple imputation by chained equations (MICE), synthetic minority oversampling technique (SMOTE), and principal component analysis (PCA). Secondly, the maximum number of iterations, maximum depth of the tree, and learning rate of the XGBoost model were optimized through WOA, and the prediction results of the model were comprehensively evaluated using accuracy, precision, recall, F1 score, and Cohen Kappa coefficient. Finally, the model was applied to predict the rock burst intensity level of the Qinlingzhongnanshan highway tunnel and the water diversion system for hydropower stations. Results show that the WOA-optimized XGBoost model achieves optimal performance when the maximum number of iterations, maximum tree depth, and learning rate are 51, 13, and 0.7325, respectively. Prediction results for rock burst intensity level using the WOA-XGBoost model outperform those of other intelligent algorithm models, verifying the model’s high accuracy and reliability in predicting rock burst intensity level.

This study investigates personnel evacuation path planning in underground the coal mine fire scenarios, with a mine in Shanxi Province, as a case study. The height, temperature, visibility and CO concentration of the smoke layer under two kinds of mine fire scenarios (Case 1 and Case 2) were numerically simulated and analyzed using FDS, a fire dynamics tool. Roadway passability coefficients was coupled with smoke hazard levels to establish the equivalent length calculation model for roadways. Incorporating smoke-related safety factors on personnel evacuation, the optimal evacuation paths were planned using the Dijkstra algorithm. Finally, evacuation simulations were performed using the Pathfinder software to validate the planned routes. Results show that: In two fire scenarios, the equivalent length of the optimal evacuation route and the shortest evacuation time for each working face have been quantified. In Case 1, the equivalent lengths of the optimal evacuation paths for each working face were 3298.8 m and 956.9 m, with the shortest evacuation time being 1595.8 s and 405.8 s, respectively. In Case 2, the corresponding equivalent lengths were 3927.2 m and 2332.4 m, with the shortest evacuation time of 1364.8 s and 786.8 s. The calculated optimal evacuation paths match the Pathfinder simulation results, confirming the validity of this optimal evacuation path planning approach based on the equivalent length of the roadway. This method could be applied to the planning of evacuation paths in coal mine fires, which provides a theoretically innovative and engineering-applicable solution for the emergency evacuation in complex mine fire environments.