Scholarlink

<p>Factors such as ground stress, gas, and coal properties cause coal and gas outbursts, which pose major safety threats during tunnel excavation. This study proposes a framework integrating the Synthetic Minority Oversampling Technique (SMOTE) algorithm with the genetic algorithm-optimized BP neural network (GA-BP). Initially, seven coupling factors governing gas outbursts in tunnels were systematically identified based on outburst disaster mechanisms. An original dataset of 53 gas tunnel samples was collected and then augmented via the SMOTE algorithm to address data imbalance. Pearson correlation analysis revealed distinct relationships between factors and outburst risk: coal seam gas pressure (+ 0.72) and coal seam gas content (+ 0.64) exhibited strong positive correlations as promoting factors (PFs), while the coal firmness coefficient (-0.53) emerged as a suppressing factor (SF). Then, the GA-BP neural network was trained using the augmented dataset to achieve 92.22% prediction accuracy. Field validations across five gas tunnels demonstrated the framework’s robustness, with hybrid measures reducing gas outburst risk to Level I (non-outburst). This work advances outburst risk management by providing actionable guidelines for dynamic risk control during gas tunnel excavation.</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

Hybrid SMOTE-GA-BP Neural Network-Based Prediction of Gas Outburst Risk in Coal-Seam-Crossing Tunnel Excavation

  • 期刊论文

  • 发表日期：

    2026年1月10日

查看全文
  • Fangzheng Ma,
  • Fei Huang,
  • Yu Zhang,
  • Yafei Luo,
  • Pengfei Wang

摘要

Factors such as ground stress, gas, and coal properties cause coal and gas outbursts, which pose major safety threats during tunnel excavation. This study proposes a framework integrating the Synthetic Minority Oversampling Technique (SMOTE) algorithm with the genetic algorithm-optimized BP neural network (GA-BP). Initially, seven coupling factors governing gas outbursts in tunnels were systematically identified based on outburst disaster mechanisms. An original dataset of 53 gas tunnel samples was collected and then augmented via the SMOTE algorithm to address data imbalance. Pearson correlation analysis revealed distinct relationships between factors and outburst risk: coal seam gas pressure (+ 0.72) and coal seam gas content (+ 0.64) exhibited strong positive correlations as promoting factors (PFs), while the coal firmness coefficient (-0.53) emerged as a suppressing factor (SF). Then, the GA-BP neural network was trained using the augmented dataset to achieve 92.22% prediction accuracy. Field validations across five gas tunnels demonstrated the framework’s robustness, with hybrid measures reducing gas outburst risk to Level I (non-outburst). This work advances outburst risk management by providing actionable guidelines for dynamic risk control during gas tunnel excavation.