Scholarlink

<p>It is extremely challenging to predict the location of rock bursts during deep underground engineering construction. Here we examined the feasibility and reliability of machine learning algorithms in rockburst location prediction, where real-time microseismic (MS) monitoring data were used. The Qinling water conveyance tunnel of the Hanjiang-to-Weihe River Diversion Project was taken as the engineering case. Firstly, the locations of MS events were directly used to forecast the location of rockbursts. The coordinate sequence of MS events of the day and the rockburst position of the following day were applied as the input and output of the neural network, respectively. Two intelligent models for the prediction of rockburst location based on the one-dimensional convolutional neural network (1D CNN) and fully connected neural network (FCNN) were then built and evaluated. The results show the prediction performance of the 1D CNN model is generally higher compared to the FCNN model. Furthermore, the spatial distribution feature of MS activities was also applied for predicting rockburst locations. The center point, radius, spatial distribution variance and energy distribution variance of MS spatial distribution were regarded as the input of prediction models. Five intelligent models for predicting rockburst location were constructed by introducing five machine learning algorithms, including adaptive boosting (Adaboost), extreme gradient boosting (XGBoost), random forest (RF), support vector machine (SVM) and gradient boosting regression tree (GBRT) algorithms. After comparing the prediction results of seven intelligent models, it is found that the prediction performance of the five models through the spatial distribution feature of MS events is much better compared to the two models using direct MS locations. Finally, an SGA model was constructed using the ensemble averaging method based on the best-performing SVM, GBRT and Adaboost. The average error of the SGA model was 2.676&#xa0;m. These findings in this study facilitate the identification and prevention of rockburst disasters.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development and Validation of Intelligent Models for Predicting Rockburst Location Based on Microseismic Monitoring and Machine Learning

  • 期刊论文

  • 发表日期：

    2025年1月26日

查看全文
  • Ke Ma,
  • Qingqing Shen,
  • Zhenghu Zhang,
  • Tao Wang

摘要

It is extremely challenging to predict the location of rock bursts during deep underground engineering construction. Here we examined the feasibility and reliability of machine learning algorithms in rockburst location prediction, where real-time microseismic (MS) monitoring data were used. The Qinling water conveyance tunnel of the Hanjiang-to-Weihe River Diversion Project was taken as the engineering case. Firstly, the locations of MS events were directly used to forecast the location of rockbursts. The coordinate sequence of MS events of the day and the rockburst position of the following day were applied as the input and output of the neural network, respectively. Two intelligent models for the prediction of rockburst location based on the one-dimensional convolutional neural network (1D CNN) and fully connected neural network (FCNN) were then built and evaluated. The results show the prediction performance of the 1D CNN model is generally higher compared to the FCNN model. Furthermore, the spatial distribution feature of MS activities was also applied for predicting rockburst locations. The center point, radius, spatial distribution variance and energy distribution variance of MS spatial distribution were regarded as the input of prediction models. Five intelligent models for predicting rockburst location were constructed by introducing five machine learning algorithms, including adaptive boosting (Adaboost), extreme gradient boosting (XGBoost), random forest (RF), support vector machine (SVM) and gradient boosting regression tree (GBRT) algorithms. After comparing the prediction results of seven intelligent models, it is found that the prediction performance of the five models through the spatial distribution feature of MS events is much better compared to the two models using direct MS locations. Finally, an SGA model was constructed using the ensemble averaging method based on the best-performing SVM, GBRT and Adaboost. The average error of the SGA model was 2.676 m. These findings in this study facilitate the identification and prevention of rockburst disasters.