A Research On Autonomous Recognition Of Source Type Based On Seismic Waveform

In the view of whole Earth, earthquakes happen every day. According to the statistics of earthquake recordings, the numbers of occurrences of earthquakes at Richter scale4,5,6and7magnitudes are respectively more than ten thousands, one thousand, one hundred and ten times annually. It may be hazardous and even incur some disasters while Richter scale is larger than6magnitudes. With the increasing intensity of human activities, the observed seismic events which originate from the explosion are also growing. The small earthquakes may be the precursor of a strong and destructive earthquake. Therefore, it is very meaningful that the event source type can be distinguished between earthquake and explosion by seismic waveform. Every day, a huge mass of observational waveform data is recorded and accumulated——such as in2007, the waveform data which are only from the seismic network of the capital Beijing has recorded increase by1.7GB every day. If the decisions of event source type for a large of seismic waveforms which are ever coming and increasing on daily routine by manpower, it needs the working staffs have rich experience and must uninterruptedly observe and analyze data. So it is almost impossible if requiring no remarkable mistake happen. Hence, the research on autonomous recognition of event source type based on seismic waveform can accurately, rapidly and timely recognize event source type in a large accumulation of seismic wave.Seismic signal is non-stationary non-linear time-variant signal. The signals collected by Seismic recorder have been interfaced by the surrounding environment and instrument own factors. In order to easily identify the event source type, it is necessary to deal with seismic waveform. Firstly, the paper eliminates the trend of the seismic signal to ensure that the signal is near the abscissa fluctuations. Secondly, it uses the difference ratio method and the autocorrelation characteristics selecting the seismic waves. Thirdly, it uses the improved wavelet threshold method to de-noising. Fourthly, the seismic signal is processed by using empirical mode decomposition (EMD) with Hilbert transformation (HT).The instantaneous amplitudes and frequencies of intrinsic mode function (IMF) which are calculated. Then, three features (the mean of autocorrelation function, the variance of cepstrum, the amplitude ratio of seismic signal S wave and P wave) are extracted form the first three IMF components. Finally, it uses non-linear separable support vector machine (SVM) to identify the type of an unknown event.This article implements autonomous recognition of event source type based on seismic waveform in MATLAB. The experiment results show that the seismic signals collected by the earthquake recorders can be judged by the two selected analysis ways to elected seismic waveform. The less seismic signal is disrupted, the more autocorrelation waveform curve tends to triangular. Experiments also demonstrated that using an improved wavelet threshold method can overcome the shortcomings of the ordinary wavelet threshold de-noising, which serious interference signal gets poor de-noised result. After using the support vector identification, the reorganization rate which used improved wavelet threshold de-noising is higher than the recognition rate which used ordinary wavelet threshold de-noising. The features are extracted from the seismic signal after Hilbert Huang Transform (HHT). The three features extracted from the first three IMF components with Hilbert Transform is recognized by SVM.The recognition rate of first IMF component with Hilbert Transform is higher about5%than the others...