Risk assessment and prediction of harmful gases in underground caverns based on TPE-XGBoost

Abstract

The risk assessment of harmful gases during the construction of underground caverns is an important component of construction safety management. However, due to the combined influence of multiple factors such as self diffusion, rock type, selection of construction machinery, ventilation scheme, etc., the risk assessment of harmful gases in caverns is often difficult to carry out. Based on this, big data monitoring of harmful gases and their influencing factors during tunnel construction was used, taking into account the multidimensional and nonlinear characteristics of the data. The eXtreme Gradient Boosting(XGBoost) ensemble learning algorithm is adopted to construct a risk assessment model for harmful gases in underground tunnels, and the Tree-structured Parzen Estimator(TPE) algorithm is used to optimize the hyperparameters of the model. By combining examples, the risk assessment result of harmful gases were predicted and the accuracy of the model was verified. The research result showed that the XGBoost model achieved an accuracy rate, recall rate, and F1 score of 85.8% for the risk assessment of harmful gases in underground caverns, demonstrating the model's accurate predictive ability. Compared with XGBoost, SVM, Decision Tree, and AdaBoost, the accuracy of the XGBoost model optimized by hyperparameters has been improved by 6.8%, 54.0%, 10.3%, and 100.9%, respectively. This proves that the model can more effectively implement the risk assessment of harmful gases in underground caverns, providing theoretical and technical guidance for construction safety management.

Keywords

underground tunnel construction / harmful gas concentration / risk assessment / XGBoost / TPE