基于信号时频特征的微震波形识别在岩爆预测中的应用
Application of microseismic waveform recognition based on signal time-frequency characteristics in rock burst prediction
投稿时间:2021-12-07  修订日期:2022-07-05
DOI:
中文关键词:  地下工程  微震监测  波形识别  频域  卷积神经网络
英文关键词:Underground engineering  Microseismic monitoring  Waveform recognition  Frequency domain  Convolutional neural network
基金项目:陕西省引汉济渭工程建设有限公司科研项目经费资助项目人工智能在引汉济渭工程岩爆预测中的应用研究
作者单位邮编
李文旭 西安理工大学 省部共建西北旱区生态水利国家重点实验室 710048
陈祖煜 中国水利水电科学研究院 100048
唐春安 大连理工大学土木工程学院 
苏 岩 陕西省引汉济渭工程建设有限公司 
唐烈先 辽宁科技大学矿业学院 
胡 晶 中国水利水电科学研究院 
陶 磊 陕西省引汉济渭工程建设有限公司 
摘要点击次数: 1110
全文下载次数: 0
中文摘要:
      在深埋地下工程施工中,需要通过监测采集微震信号分析施工过程中的岩爆风险,由于现场干扰因素多,数据中混入了大量冗余信号,大大影响了岩爆预测的效率。为有效识别出岩石的破裂信号,本文采用快速傅里叶变换,对比分析了微震/岩爆信号与其它无效信号的频率特征,采用多输入的卷积神经网络方法,建立了基于信号时频特征的微震波形识别模型,实现了对微震/岩爆信号的有效识别。基于引汉济渭秦岭输水隧洞的微震监测资料,采用3770个波形对模型进行了测试,模型识别精度可达96.1%。模型对比了不同输入方式对预测结果的影响,针对随机挑选的100次微震事件和100次无效事件,结果表明:采用信号的时频特征作为输入,模型比单纯采用时域或频域特征具有更高的精度。
英文摘要:
      In the construction of deep-buried underground engineering, it is necessary to analyze the rock burst risk in the construction process by monitoring and collecting microseismic signals. Due to the multiple interference factors on site, a large number of redundant signals are mixed into the data, which greatly affects the efficiency of rock burst prediction. To effectively identify the rock burst signal, this paper uses fast Fourier transform, analyzing frequency characteristics of microtremor/rock burst with other invalid signals, using multiple input convolutional neural network method, based on the signal time-frequency characteristics of the micro wave shape recognition model, implements the effective identification of microtremor/rock burst signal. Based on the microseismic monitoring data of Qinling water conveyance tunnel of Hanjiang to Weihe River valley water diversion project, 3770 waveforms are used to test the model, and the model recognition accuracy can reach 96.1%. The model compares the influence of different input methods on the prediction results, and for the randomly selected 100 microseismic events and 100 invalid events, the results show that the model with the time-frequency characteristics of the signal as input has higher accuracy than that with the time-domain or frequency-domain characteristics.
  查看/发表评论  下载PDF阅读器
关闭