Scholarlink

<p>The coal burst hazard varies across different regions of a mining face, necessitating the development of an effective spatial hazard early warning method to enable timely and targeted preventive measures. This study proposes a spatial early warning method for coal burst that integrates the spatial extraction of multi-parameter warning information with machine learning, enabling the early prediction of coal burst hazard levels at different locations across the mining face. This method divides the mining face into grids and utilizes a sliding window constructed from these grids to enable spatial data scanning. Based on microseismic monitoring, it accurately extracts warning information and corresponding hazard levels from spatial grids. Suitable indicator combinations are selected according to the early warning effectiveness for high-energy events, forming a multi-parameter warning dataset. The hyperparameters of the Extreme Gradient Boosting (XGBoost) network are efficiently optimized using Bayesian optimization, and five-fold cross-validation is introduced to enhance the model’s generalization capability. A spatial early warning model for coal burst hazards (Bayes-XGBoost) is developed, enabling detailed hazard predictions by region and level for the mining face. A case study was conducted using field data from a mining face.</p><p>The results indicate that the Bayes-XGBoost model, after being trained on a sufficient spatial early warning dataset, can identify the coal burst hazard levels of specific regions. The microseismic event energy level-frequency distribution demonstrates staged characteristics, including upward trends, linear declines, increased dispersion, and oscillation, which allow for the determination of critical energy levels and, in turn, the classification of hazard levels. An optimal combination of indicators responsive to high-energy events was identified, and the spatial dataset constructed with these indicators provided effective input for the early warning model. This model achieved prediction accuracies of 91.2% and 88.9% for energy level 5 events, 1 day and 2 days in advance, respectively. For events at an energy level of 4, prediction accuracies of 79.0% and 76.5% were achieved 1 day and 2 day in advance. Balancing energy level and lead time ensures that prediction accuracy remains within an acceptable range. This study provides a novel approach for spatial early warning of coal burst hazards.</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

The Spatial Hazard Early Warning Method for Coal Burst Based on Fusion of Multi-parameter Indicators

  • 期刊论文

  • 发表日期：

    2025年10月6日

查看全文
  • Shifeng He,
  • Heping Xie,
  • Minghui Li,
  • Feng Cui,
  • Jingyang Zhang,
  • Jun Lu,
  • Farui Shi

摘要

The coal burst hazard varies across different regions of a mining face, necessitating the development of an effective spatial hazard early warning method to enable timely and targeted preventive measures. This study proposes a spatial early warning method for coal burst that integrates the spatial extraction of multi-parameter warning information with machine learning, enabling the early prediction of coal burst hazard levels at different locations across the mining face. This method divides the mining face into grids and utilizes a sliding window constructed from these grids to enable spatial data scanning. Based on microseismic monitoring, it accurately extracts warning information and corresponding hazard levels from spatial grids. Suitable indicator combinations are selected according to the early warning effectiveness for high-energy events, forming a multi-parameter warning dataset. The hyperparameters of the Extreme Gradient Boosting (XGBoost) network are efficiently optimized using Bayesian optimization, and five-fold cross-validation is introduced to enhance the model’s generalization capability. A spatial early warning model for coal burst hazards (Bayes-XGBoost) is developed, enabling detailed hazard predictions by region and level for the mining face. A case study was conducted using field data from a mining face.

The results indicate that the Bayes-XGBoost model, after being trained on a sufficient spatial early warning dataset, can identify the coal burst hazard levels of specific regions. The microseismic event energy level-frequency distribution demonstrates staged characteristics, including upward trends, linear declines, increased dispersion, and oscillation, which allow for the determination of critical energy levels and, in turn, the classification of hazard levels. An optimal combination of indicators responsive to high-energy events was identified, and the spatial dataset constructed with these indicators provided effective input for the early warning model. This model achieved prediction accuracies of 91.2% and 88.9% for energy level 5 events, 1 day and 2 days in advance, respectively. For events at an energy level of 4, prediction accuracies of 79.0% and 76.5% were achieved 1 day and 2 day in advance. Balancing energy level and lead time ensures that prediction accuracy remains within an acceptable range. This study provides a novel approach for spatial early warning of coal burst hazards.