基于经验模态分解的地震波特征提取的研究

Study on seismic signal features extraction based on EMD

本文原始数据为35个天然地震和27个人工爆破事件的离震中最近的5个台站的垂直分量波形数据.从原始波形数据分别提取最大振幅对应的周期、倒谱的方差、自相关函数的最大值3个特征.依据希尔伯特-黄变换原理,用经验模态分解方法把原始波形信号分解为10个左右的本征模态函数分量后,再从每个分量中分别提取这3个特征.接着对所获取的特征样本集合采用随机划分法分为学习样本集与检验样本集,然后再通过支持向量机进行分类识别,如此反复进行多次样本划分和分类识别.结果表明经验模态分解后的分量信号提取的这3个特征具有更高的识别率,说明了经验模态分解有利于识别天然地震和人工爆破事件,值得进一步深入研究.

The research on seismic signal processing,analysis,and further discrimination of earthquakes and explosions plays a fundamental role in the development of seismology and other geosciences.And explosion event recognition,especially nuclear explosion recognition,is indispensable for public welfare and world peace.This paper introduces some important results of our research project：＂Classification and Decision Support System for Recognizing Natural Earthquake Events and Explosion Events＂.The selected raw data are the up-down components of 5 observatory stations which are the nearest ones to the epicenters of 35 earthquake events occurred during the years of 2003~2007 and 27 explosion events occurred during the years of 2002~2008.From the raw waveform data,3 features：ratio of the maximal amplitude of S-Wave and P-Wave,cepstral variance,maximum of autocorrelation,are extracted.And based upon Hilbert-Huang Transform（HHT） principle,by means of empirical mode decomposition（EMD） method,original waves are decomposed into about 10 intrinsic mode functions（IMFs）.These 3 features are also extracted from each of the IMFs for every selected wave signals.The acquired about features samples are randomly partitioned into training sample set and testing sample set.Finally,Support vector machine（SVM） classifier for discriminating earthquake and explosion events are built and trained by training sample set and tested by testing sample set.Such a process of samples-partition,training,testing are reiterated 100 times.The results show that these 3 features extracted from IMFs after EMD is better than those extracted from raw signal with higher correct recognition rate.This implies EMD processing of seismic wave is beneficial for recognition of earthquake and explosion events and deserves more intensive research.