物理指标与深度学习融合的冲击地压风险等级预测  

Prediction of rock burst risk level based on combination of physical indexes and deep learning

作　　者：乔美英[1,2] 史有强 QIAO Meiying;SHI Youqiang(School of Electrical Engineering and Automation,Henan Polytechnic University,Jiaozuo Henan 454000,China;Henan International Joint Laboratory of Direct Drive and Control of Intelligent Equipment,Jiaozuo Henan 454003,China)

机构地区：[1]河南理工大学电气工程与自动化学院,河南焦作454000 [2]河南省智能装备直驱技术与控制国际联合实验室,河南焦作454003

出　　处：《中国安全生产科学技术》2024年第4期56-63,共8页Journal of Safety Science and Technology

基　　金：国家自然科学基金项目(U1404510);河南省科技攻关项目(222102220076)。

摘　　要：为探究煤矿智能化开采背景下,冲击地压的预警问题。以数据分析为基础,以河南某矿21181工作面为背景,提出物理指标与深度学习融合的冲击地压预警方法。该方法通过分析大能量事件发生之前,各项物理指标在最大值、趋势性以及相对变化率绝对值3方面特性,得到与之对应的综合物理特征,并分析震源的空间分布特性,根据其特性提出坐标注意力机制,对震源坐标进行加权,得到震源特征。对综合物理特征以及震源特征,加入通道注意力机制对特征进行加权,并使用全连接层进行分类,达到风险等级预测的目的,最终将模型运用于实际工程。研究结果表明:物理指标与深度学习融合的冲击地压预警方法可以达到较高的准确率。研究结果可为实际工程提供一定借鉴。In order to explore the early-warning problem of rock burst under the background of intelligent mining of coal mine,taking 21181 working face of Qianqiu Mine in Henan Province as the background,an early-warning method of rock burst combining the physical indexes and deep learning was proposed based on data analysis.By analyzing the characteristics of the maximum value,trend and absolute value of relative change rate of each physical index before the occurrence of the big energy event,the corresponding comprehensive physical characteristics were obtained.The spatial distribution characteristics of seismic source were analyzed,then the coordinate attention mechanism was proposed according to their characteristics,and the seismic source coordinates were weighted to obtain the seismic source characteristics.The comprehensive physical characteristics and seismic source characteristics were weighted by adding the channel attention mechanism.The whole connection layer was used for classification to achieve the purpose of risk grade prediction,and finally the model was applied to the actual project.The results show that the early-warning method of rock burst combining the physical indexes and deep learning can achieve a high accuracy,and the research results can provide some reference for practical engineering.

关 键 词：冲击地压 物理指标 数据分析 震源特征 深度学习 坐标注意力机制 

分 类 号：X936[环境科学与工程—安全科学]