Scholarlink

<p>Underground mining in hard-rock mines is conducted under complex geological conditions, where rock masses are prone to sudden failure. To address this challenge, this study proposes an integrated early-warning framework coupling microseismic (MS) monitoring with a Convolutional Neural Network – Long Short-Term Memory (CNN-LSTM) predictor and a Cumulative Sum (CUSUM) change-point detector. The CNN-LSTM forecasts key MS parameters (seismic energy and source radius) from real-time monitoring data. A new indicator, <i>Q</i>—defined as the ratio of seismic energy to the cube of source radius—is introduced to quantify energy release density during rock fracturing. CUSUM tracks the evolving <i>Q</i> series to identify significant deviations and issue early warnings. The framework is validated using MS data from the Dongtangzi lead – zinc mine in the tectonically active Qinling Orogenic Belt, northwestern China. Results show that, compared with conventional approaches, prediction performance is markedly improved: relative to a standalone LSTM, RMSE and MAE decrease by ∼30% – 56%, while <i>R</i><sup>2</sup> increases by ∼20% – 58%. The framework remains robust under complex monitoring conditions and provides warning signals up to one week in advance. Moreover, the post-blasting attenuation of <i>Q</i> provides quantitative insights into stress-release mechanisms and surrounding rock behavior. These findings highlight the framework’s practical applicability and theoretical significance for mining hazard prediction.</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

Temporal analysis and early warning of rock instability in hard rock mines based on CNN-LSTM and CUSUM

  • 期刊论文

  • 发表日期：

    2026年4月27日

查看全文
  • Fang Yan,
  • Guan-guan Li,
  • Long-jun Dong,
  • Sheng-nan Du,
  • Fei-fan He

摘要

Underground mining in hard-rock mines is conducted under complex geological conditions, where rock masses are prone to sudden failure. To address this challenge, this study proposes an integrated early-warning framework coupling microseismic (MS) monitoring with a Convolutional Neural Network – Long Short-Term Memory (CNN-LSTM) predictor and a Cumulative Sum (CUSUM) change-point detector. The CNN-LSTM forecasts key MS parameters (seismic energy and source radius) from real-time monitoring data. A new indicator, Q—defined as the ratio of seismic energy to the cube of source radius—is introduced to quantify energy release density during rock fracturing. CUSUM tracks the evolving Q series to identify significant deviations and issue early warnings. The framework is validated using MS data from the Dongtangzi lead – zinc mine in the tectonically active Qinling Orogenic Belt, northwestern China. Results show that, compared with conventional approaches, prediction performance is markedly improved: relative to a standalone LSTM, RMSE and MAE decrease by ∼30% – 56%, while R2 increases by ∼20% – 58%. The framework remains robust under complex monitoring conditions and provides warning signals up to one week in advance. Moreover, the post-blasting attenuation of Q provides quantitative insights into stress-release mechanisms and surrounding rock behavior. These findings highlight the framework’s practical applicability and theoretical significance for mining hazard prediction.