Researches On The Algorithms Of Automatic Segmentation Of Seismic Signals And The Recognition Of Source Types

The seismic event is an inevitable event in the process of the earth's evolution,and the seismic activity has the great potential to be extremely destructive.Strong earthquake events in densely populated areas will cause great damage to people's lives and property.Thus seismic monitoring network has been established to monitor seismic activity,and the seismic waveform signal is the only way to monitor seismic activity.Many non-natural earthquake events such as chemical explosion,underground nuclear explosion,landslides,mine collapse,ground subsidence,and even large construction active can produce seismic wave and propagate energy to the surroundings,with the increase of observation network layout and intensive observation instruments and improved sensitivity,as the result,all kinds of seismic waves will be detected by seismic observation instrument.For the natural earthquake,there has three key elements:source,magnitude and occurrence time,it can be obtained from the network observation waveform;of course,the characteristics of seismic waveform inversion propagation path is the only way to explore the internal structure of the earth.From the continuous observation signal automatic interception of a seismic waveform signal,corresponding to the effective event source and seismic waveform signal and the extracted features to identify the source type(the thesis only focuses on earthquake and explosion classification)is the main content of this study.In this paper,3 problems are discussed,such as the identification of the source types,the seismic signal P wave and the S wave first arrival time detection and automatic waveform extraction,the wave group(noise,P wave group and S wave group)identification.In this paper,a new algorithm for identifying the source types based on BP-Adaboost method is proposed.The BP neural network is selected as the weak classifier,and using the Adaboost method combined into a strong classifier.Compared with BP neural network and PCA-SVM method for earthquake and explosion classification in the real datasets,the results show that over 98%classification rate has obtained by BP-Adaboost,and with the good generalization ability.This paper proposes an algorithm for rapid and automatic detection of the arrival time of P-wave and S-wave from seismic waveform signals:firstly,after the recorded whole waveform signal which covering the full process of an earthquake events filtered as a whole to eliminated noticeable noises,the filtered seismic signal is segmented for many wave fragments with a window length of 64 sampling points for each wave fragment;after that,5 layers EMD decomposition is carried out on each wave fragment respectively,the corresponding IMFs are got;then the TKEO energies are calculated from the IMF1 components which being decomposed by EMD from all wave fragment respectively,and also normalized;finally,some suitable energy threshold is chosen to infer the arrival times of P-wave and the S-wave.The experimental results show that compared with the state of art STA-LTA and AR-AIC method,this method has a significantly shorter computation time,and can identify the arrival time of P wave earlier.By setting a proper threshold,also can extract the whole effective seismic waveform covers the whole process of an event signal,so that use this method can remove part of the non-continuous observation event signal.The experiment on seismic data show that the average compression ratio reaches 57%,which can significantly reduce the seismic signal storage space.The recognition criterion of detection the wave group is proposed:the average spectral energy value and STD.Based on the EMD-TKEO algorithm,the accurate extraction of noise,P wave and S wave group sample can be got,then use the expended multi-classification SVM method get the classification result.The value obtained(C,?)by grid search of the tuple(1,0.01)to identify the sample wave group was the best,the identification rate is 99%,with the high efficiency,the experiments also show that the effectiveness of the proposed wave group features,just use a few features can achieve the ideal recognition rate.