Scholarlink

<p>Accurate prediction of short-term rockbursts is hindered by the quality of micro-seismic monitoring data, which often suffers from missing values, outliers, and class imbalance. To address these challenges, this study proposes a customized processing framework for poor-quality data in short-term rockburst prediction integrating K-nearest neighbors (KNN) imputation, cluster-based local outlier factor (CBLOF) filtering, and support vector machine synthetic minority oversampling technique (SVMSMOTE). Based on a database of 132 rockburst samples constructed from engineering projects, six ensemble learning models were trained, with hyperparameters optimized using the Gray Wolf Optimizer (GWO). Quantitative results indicate significant performance gains: the optimized GWO-CS-RF and GWO-CS-AdaBoost models achieved the highest accuracy improvements of 15% over their baseline counterparts, demonstrating the effectiveness of the proposed data quality enhancement strategy. Interpretability analysis using SHAP identified cumulative micro-seismic energy <i>E</i> as the most critical precursor feature for rockburst intensity. Furthermore, the proposed model was validated across three distinct engineering projects in Asia, achieving an overall prediction accuracy of 94.4%. Improving data quality is often more effective than merely increasing model complexity. Consequently, the customized data processing framework proposed in this study offers a robust solution for enhancing the accuracy of short-term rockburst early warnings.</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

Application of Data Processing Methods and Optimization Algorithms Based on Micro-seismic Monitoring in Short-Term Rockburst Prediction

  • 期刊论文

  • 发表日期：

    2026年4月15日

查看全文
  • Weiqiu Kong,
  • Peng Hou,
  • Quansheng Liu,
  • Zhijun Wu,
  • Mukhammedrakhym Rabatuly,
  • Xin Liang

摘要

Accurate prediction of short-term rockbursts is hindered by the quality of micro-seismic monitoring data, which often suffers from missing values, outliers, and class imbalance. To address these challenges, this study proposes a customized processing framework for poor-quality data in short-term rockburst prediction integrating K-nearest neighbors (KNN) imputation, cluster-based local outlier factor (CBLOF) filtering, and support vector machine synthetic minority oversampling technique (SVMSMOTE). Based on a database of 132 rockburst samples constructed from engineering projects, six ensemble learning models were trained, with hyperparameters optimized using the Gray Wolf Optimizer (GWO). Quantitative results indicate significant performance gains: the optimized GWO-CS-RF and GWO-CS-AdaBoost models achieved the highest accuracy improvements of 15% over their baseline counterparts, demonstrating the effectiveness of the proposed data quality enhancement strategy. Interpretability analysis using SHAP identified cumulative micro-seismic energy E as the most critical precursor feature for rockburst intensity. Furthermore, the proposed model was validated across three distinct engineering projects in Asia, achieving an overall prediction accuracy of 94.4%. Improving data quality is often more effective than merely increasing model complexity. Consequently, the customized data processing framework proposed in this study offers a robust solution for enhancing the accuracy of short-term rockburst early warnings.