In order to realize the effective judgment of surrounding rock classification under the condition of imbalanced surrounding rock classification and improve the simulation accuracy of tunnel construction, a study on the tunnel construction simulation is carried out based on the advanced classification of imbalanced surrounding rock with improved eXtreme Gradient Boosting (XGBoost). The Automatic Neighborhood size Determination-SMOTE (AND-SMOTE) method is used to optimize the class imbalance of surrounding rock, and the improved XGBoost model is used for advanced classification of surrounding rock, and then the simulation parameters are optimized, which improved the accuracy of tunnel construction simulation. In order to improve the accuracy of surrounding rock classification, the Harris Hawks Optimization (HHO) algorithm is used to automatically optimize the hyperparameters of XGBoost ensemble classifier. The engineering application shows that the improved XGBoost model has higher classification accuracy than the unimproved XGBoost, KNN, SVC and other 6 models. The classification accuracy of the improved XGBoost model is improved by 8.6% after considering the class imbalance of surrounding rock. In addition, the relative deviation between the results of tunnel construction simulation based on the advanced classification of surrounding rock and the actual progress is reduced by 11.3% compared with the traditional simulation, which is more in line with the engineering reality.