Scholarlink

<p>Historical early warning text information of coal mine major disasters has important reference value for decision-making and deployment when facing danger. Although most coal mines are equipped with a comprehensive safety monitoring system to achieve early warning of potential disasters, the lack of real-time and diversified early warning information mining capabilities cannot provide effective support for decision-making. This study utilized text mining technology based on a pre-trained language model to identify key information for coal mine disaster early warnings. First, 562 accident investigation reports from 2003 to 2024 were collected and 4781 textual data entries were compiled. Disaster text was analyzed using the latent Dirichlet allocation model to extract disaster topic and feature item weights. Subsequently, the text was segmented to construct major disaster information databases, considering the frequency of keywords. Finally, the preprocessed text was input into the BERT pre-trained language model for fine-tuning to extract five key information, five general disaster types and thirteen specific disaster types. A case study demonstrated that the optimized BERT model combined with sliding window dynamically identified early warning key information, including disaster type, time, location, causes and measures, with an accuracy rate of 96.85%, which provides robust guidance for real-time, efficient decision-making in response to major coal mine disasters and ensures the safety and reliability of production.</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

Text Mining and Early Warning Key Information Identification of Coal Mine Major Disasters Utilizing Pre-trained Language Model

  • 期刊论文

  • 发表日期：

    2025年10月13日

查看全文
  • Shenglei Zhao,
  • Enyuan Wang,
  • Zhonghui Li,
  • Jinxin Wang,
  • Tingjiang Tan,
  • Yubing Liu,
  • Qiming Zhang,
  • Baolin Li

摘要

Historical early warning text information of coal mine major disasters has important reference value for decision-making and deployment when facing danger. Although most coal mines are equipped with a comprehensive safety monitoring system to achieve early warning of potential disasters, the lack of real-time and diversified early warning information mining capabilities cannot provide effective support for decision-making. This study utilized text mining technology based on a pre-trained language model to identify key information for coal mine disaster early warnings. First, 562 accident investigation reports from 2003 to 2024 were collected and 4781 textual data entries were compiled. Disaster text was analyzed using the latent Dirichlet allocation model to extract disaster topic and feature item weights. Subsequently, the text was segmented to construct major disaster information databases, considering the frequency of keywords. Finally, the preprocessed text was input into the BERT pre-trained language model for fine-tuning to extract five key information, five general disaster types and thirteen specific disaster types. A case study demonstrated that the optimized BERT model combined with sliding window dynamically identified early warning key information, including disaster type, time, location, causes and measures, with an accuracy rate of 96.85%, which provides robust guidance for real-time, efficient decision-making in response to major coal mine disasters and ensures the safety and reliability of production.