Scholarlink

<p>During the construction of railway tunnels in western China, increased excavation depths led to challenges such as rockbursts. However, the complex relationship between microseismicity and rockbursts often complicates achieving accurate results, primarily due to subjective human decisions in selecting relevant inputs and classifying rockburst risks efficiently. Therefore, this paper proposes an intelligent model for the efficient classification of rockburst intensities by studying the relationship between microseismicity and rockbursts, while extracting the most significant microseismic inputs for the classifier using feature selection techniques. Initially, a database of 138 rockburst cases from underground engineering projects, containing various microseismic parameters, was established. Feature selection techniques, including filter methods, were employed to extract statistically significant microseismic features. A wrapper feature selection method was then applied to the filter method outputs to determine the optimal feature subset for the ExtraTrees (ET) classifier. The most significant parameters identified were the combination of the cumulative number of events (MN), cumulative microseismic energy (ME) and cumulative microseismic apparent volume (MV), which were used to create a predictive model by optimising hyperparameters embedding the cross-validation technique to predict rockburst risks. The results showed that for our classification problem when the model was evaluated using the performance metric F1 score, it achieved a score of 0.90, reflecting strong predictive performance when trained with a subset of these features. Moreover, the model’s global explanation was utilised to understand the link between these features and risk levels, indicating that basically lower values of ME, MN and MV suggest no rockburst risk; low to medium values imply a slight risk; medium to high values indicate a moderate risk; and high values across all parameters signal intense rockbursts. Furthermore, local explanations identified the reasons behind the misclassification of certain samples, shedding light on the complex rockburst mechanisms in specific instances. Ultimately, the model was validated with case studies from a high-altitude railway tunnel, demonstrating effective risk prediction. The proposed work can extract the relationship between microseismic parameters and rockbursts, and classify risk levels more efficiently using fewer and more relevant indicators.</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

Evaluating the Rockburst Potential in the Railway Tunnel after Identifying the Most Significant Microseismic Inputs for the Classifier using Feature Selection Techniques

  • 期刊论文

  • 发表日期：

    2025年7月30日

查看全文
  • Prabhat Man Singh Basnet,
  • Guangliang Feng,
  • Aibing Jin,
  • Jingwen Chen,
  • Xinyuan Wei,
  • Shakil Mahtab,
  • Hilal Ahmad

摘要

During the construction of railway tunnels in western China, increased excavation depths led to challenges such as rockbursts. However, the complex relationship between microseismicity and rockbursts often complicates achieving accurate results, primarily due to subjective human decisions in selecting relevant inputs and classifying rockburst risks efficiently. Therefore, this paper proposes an intelligent model for the efficient classification of rockburst intensities by studying the relationship between microseismicity and rockbursts, while extracting the most significant microseismic inputs for the classifier using feature selection techniques. Initially, a database of 138 rockburst cases from underground engineering projects, containing various microseismic parameters, was established. Feature selection techniques, including filter methods, were employed to extract statistically significant microseismic features. A wrapper feature selection method was then applied to the filter method outputs to determine the optimal feature subset for the ExtraTrees (ET) classifier. The most significant parameters identified were the combination of the cumulative number of events (MN), cumulative microseismic energy (ME) and cumulative microseismic apparent volume (MV), which were used to create a predictive model by optimising hyperparameters embedding the cross-validation technique to predict rockburst risks. The results showed that for our classification problem when the model was evaluated using the performance metric F1 score, it achieved a score of 0.90, reflecting strong predictive performance when trained with a subset of these features. Moreover, the model’s global explanation was utilised to understand the link between these features and risk levels, indicating that basically lower values of ME, MN and MV suggest no rockburst risk; low to medium values imply a slight risk; medium to high values indicate a moderate risk; and high values across all parameters signal intense rockbursts. Furthermore, local explanations identified the reasons behind the misclassification of certain samples, shedding light on the complex rockburst mechanisms in specific instances. Ultimately, the model was validated with case studies from a high-altitude railway tunnel, demonstrating effective risk prediction. The proposed work can extract the relationship between microseismic parameters and rockbursts, and classify risk levels more efficiently using fewer and more relevant indicators.