当前位置：
X-MOL 学术
›
Tunn. Undergr. Space Technol.
›
论文详情
Our official English website, www.x-mol.net, welcomes your
feedback! (Note: you will need to create a separate account there.)
Predicting short-term rockburst intensity using a weighted probability stacking model with optimal feature selection and Bayesian hidden layer
Tunnelling and Underground Space Technology ( IF 6.7 ) Pub Date : 2024-08-20 , DOI: 10.1016/j.tust.2024.106021 Jiahao Sun , Wenjie Wang , Lianku Xie
Tunnelling and Underground Space Technology ( IF 6.7 ) Pub Date : 2024-08-20 , DOI: 10.1016/j.tust.2024.106021 Jiahao Sun , Wenjie Wang , Lianku Xie
Rockburst is the most common and severe geological hazard affecting the safe construction of underground projects, which seriously endangers the security of construction personnel and facilities. This study proposes a weighted probability stacking model with optimal feature selection and Bayesian hidden layer (OFS-Bayes-WPS) to predict short-term rockburst intensity. In this regard, 114 rockburst cases with microseismic (MS) parameters were gathered to build a database for modelling. The database was split into a training set (80%) and a test set (20%) using random stratified sampling. The cumulative number of MS events, cumulative MS energy, cumulative MS apparent volume, and MS energy rate were selected as prediction indicators using a correlation-based feature selection algorithm. In the modelling process, 5-fold cross-validation and Bayesian optimization were utilized to find the optimal hyper-parameters of the nine basic models. The evaluation results show that the OFS-Bayes-WPS model reached 91.3% accuracy in the test set, whose prediction performance outperforms other single and ensemble models. Moreover, the model shows superior performance compared with previously proposed machine learning models for rockburst prediction. For further verifying the applicability and practicability, OFS-Bayes-WPS was utilized to predict five new rockburst cases in the Ashele Copper Mine in China, the Neelum–Jhelum Hydroelectric Tunnel in Pakistan and the Qinling Water Conveyance Tunnel in China, and it showed excellent performance. The predictive advantages of OFS-Bayes-WPS in slight and moderate rockburst are revealed by comparing the changes in the performance of the model before and after introducing the Bayesian hidden layer. In addition, the impact of the optimal feature selection mechanism on stacking model training efficiency is analyzed. When the number of basic learners and training samples is large, the optimal feature selection can effectively improve the model training efficiency and decrease the risk of overfitting while keeping accuracy. The method can offer a proven guide for short-term rockburst prediction.
更新日期：2024-08-20
京公网安备 11010802027423号