Researches On EMD For Identifying Seismic Source Types

Empirical mode decomposition(EMD) method has obvious advantages for processing non-stationary and nonlinear signals. Different from traditional signal processing methods, the EMD method has obvious advantages in analyzing the local features of signals and can truly reflect the signal energy distribution in time scale and frequency scale. On these grounds, this thesis use a wavelet threshold based on EMD method to de-noise the seismic signal. Noise signals commonly embedded as the high frequency components. In order to de-noise the seismic signal, firstly the seismic signal needs to be decomposed by the EMD method, then the higher-frequency-embedded components-the first three intrinsic mode functions(IMF) are de-noised by wavelet threshold method, finally all IMFs are summed to compose the de-noised signal. Being compared with ordinary wavelet threshold method, the EMD-based wavelet threshold method has a higher signal-to-noise ratio, which means a better de-noising effect.This thesis focuses on the feature extraction researches of seismic signal's IMF energy ratio and singular value entropy. It is radically different that focal mechanisms of earthquakes and explosions:earthquake is shearing source and explosion is expanding source. This difference may incur obvious discrepancies in the components of the two types seismic signals' IMFs which can be expressively simplified in the form of component energy. By calculating the ratios of the energy of every IMF component and the total energy of the whole waveform signal, these ratio features may be utilized to clearly differentiate between the two different event source types. Singular value is an inherent property of matrix:if the matrix elements maintains small variability, then it's singular values is also small fluctuation; different IMF components comprise the different frequency components of the original wave signal, the singular value in each IMF reflects the distribution of signals in different frequency bands; entropy reflects this uniformity of distribution, more uniform distribution of singular value, the greater the entropy, otherwise the smaller. In this thesis, above two features -- energy ratio and entropy, are extracted from seismic signals of earthquake and explosion events. By the experiments being conducted on 35 earthquakes and 27 explosion events occurred in the Beijing-nearby area -- feature extraction by above two methods and recognition with support vector machine (SVM), it has been found that the IMFs' energy ratio and singular value entropy features can significantly distinguish between earthquakes and explosion wave event, and the discriminatory capability of IMFs' energy ratio is better than the singular value entropy.Finally, more experiments for feature extraction and recognition testing are conducted by constructed secondary-order feature based on IMFs energy ratio features using "(Bright-Lampe No.1)(Ming-Deng No.1)" exploration explosion experiment and more other 62 natural earthquake events occurred in the same area, the recognition results are also very encouraging. This thesis believes that:the IMFs energy ratio and singular value entropy can be considered as option features to discriminate earthquake and explosion.