摘要
准确分类爆破和塌陷等非构造地震是细化地震监测的重要需求.不同地区的非构造地震具有区域差异,需要针对性的开展研究.本文对内蒙古自治区的构造地震、爆破和塌陷三类事件进行了分析,利用2016—2022年在内蒙古发生的ML 1.5级以上的289个地震事件的2323条高质量垂直向波形记录,提取了拐角频率、P/S谱比等7个类别共31个人工定义的波形特征参数,分析了不同类型事件各类参数的特征.利用14个优选参数组成的特征向量,使用XGBoost方法分类地震并评估了各参数的重要性.结果表明,本文提取的7类波形特征参数能够表征地震类型差异,且爆破事件的拐角频率、波形复杂度、瞬时频率与倒谱具有明显的分区差异.基于多个特征向量的三分类判别中,XGBoost方法优于支持向量机(SVM)、长短期记忆网络(LSTM)等机器学习方法,准确率90%.基于XGBoost的权重分析表明P/S谱比(16~20 Hz)在三分类识别中权重最高.研究结果表明由人工定义的波形特征参数能够反映不同事件震源性质的差异,可通过构建分类器实现地震事件类型的判别,进而服务地震活动性分析和应急处置等实际需求.
Accurate classification of non-tectonic earthquakes,such as blasts and collapses,is a critical requirement for enhancing earthquake monitoring.Due to regional variations in non-tectonic earthquakes,targeted studies are essential.This study analyzes three types of seismic events in the Inner Mongolia Autonomous Region:tectonic earthquakes,blasts,and collapses.Using 2323 high-quality vertical waveform records from 289 seismic events(ML≥1.5)recorded between 2016 and 2022 in Inner Mongolia,a total of 31 manually defined waveform feature parameters were extracted,encompassing 7 categories such as corner frequency and P/S spectral ratio.The characteristics of these parameters for different event types were examined.A feature vector composed of 14 optimized parameters was used with the XGBoost method to classify the seismic events and assess the importance of each parameter.The results demonstrate that the seven categories of waveform features effectively capture differences among event types.Specifically,blast events exhibit distinct regional variations incorner frequency,waveform complexity,instantaneous frequency,and cepstrum.In three-category classifications,XGBoost outperforms other machine learning methods,such as SVM and LSTM,achieving an accuracy of 90%.Weight analysis using XGBoost indicates that the P/S spectral ratio(16~20 Hz)has the highest weight in classification.These findings suggest that manually defined waveform features can effectively reflect differences in source characteristics of various events.By constructing classifiers,event types can be reliably identified,supporting seismic activity analysis and emergency response efforts.
作者
张珂
王伟涛
王鑫
赵辉
ZHANG Ke;WANG WeiTao;WANG Xin;ZHAO Hui(Earthquake Bureau of Inner Mongolia Autonomous Region,Hohhot 010010,China;Institute of Geophysics,China Earthquake Administration,Beijing 100081,China;Key Laboratory of Source Physics,China Earthquake Administration,Beijing 100081,China)
出处
《地球物理学报》
北大核心
2025年第10期3849-3868,共20页
Chinese Journal of Geophysics
基金
内蒙古自治区自然科学基金(2023LHMS04006)
国家自然科学基金项目(42074060)
内蒙古自治区地震局局长基金(2024MS02)资助.
关键词
构造地震
人工爆破
塌陷
时频域特征提取
自动识别
Tectonic earthquake
Artificial blasting
Collapse
Time-frequency domain feature extraction
Automatic recognition
作者简介
第一作者:张珂,女,1993年生,硕士,工程师,2018年毕业于中国海洋大学,主要从事数字地震学方面研究.E-mail:zkee0928@163.com;通讯作者:王伟涛,男,1979年生,博士,研究员,主要从事主动震源、台阵探测和人工智能方面的研究.E-mail:wangwt@cea-igp.ac.cn。
