关键词: 地下综合管廊, 安全状况评估, XGBoost, 贝叶斯优化, 特征工程

Abstract: Concerning the complex security situation and difficult risk assessment of the underground comprehensive pipe gallery，a safety assessment method based on XGBoost algorithm was proposed，and the data of the underground comprehensive pipe gallery was used to build the model. Firstly，the data collected from the underground comprehensive pipe gallery were detected for outliers and processed for missing values，and the statistical features and cross features were constructed by the method of descriptive statistics and feature combination. Secondly，Logistic Regression（LR），Random Forest（RF），Nearest Neighbor Classifier（NC），Support Vector Machine（SVM）and XGBoost algorithm were used to build the safety evaluation model. Finally，Bayesian algorithm was used to optimize the model parameters. The experimental results show that the optimized XGBoost model has a higher accuracy in the safety assessment of underground comprehensive pipe gallery compared with LR，RF，NC，SVM，with a maximum of 0. 920 9.

Key words: utility tunnel, safety situation assessment, XGBoost, Bayesian optimization, feature engineering