| Earthquake occur at every moment.Large amounts of seismic data are produced in the world at every minute.And the seismic data have abundant and various information.How to deal with those seismic data is facing great challenges.Seismic phase picking is the primary work in the field of seismic monitoring.The P-wave is the fastest wave in the seismic wave and the first to reach the earth's surface.Picking up P-wave precisely is the key to study the seismic source location and seismic tomography,etc.The traditional manual picking method can not meet the needs of deep exploration of the earth in terms of accuracy and speed.With the rapid development of sensor network technology,the detection of seismic signals,picking of phases,and detection of P-waves is becoming less difficult.Usually,the seismic signals captured by sensor networks are time-varying data.However,due to the presence of redundant information and noise in the seismic time-varying data under different environmental conditions,the existing pick-up methods are not accurate and are depended on the empirical threshold.To improve the accuracy of picking up P-wave,this paper extracts the features of the seismic time-varing data and uses data mining method to pick up P-wave precisely.The main work and innovation as follows:1.By analyzing the seismic time-varing data comprehensively,15 dimensional features are extracted from time domain and frequency domain.One single feature does not characterize the seismic time-varing data fully.Many automatic phase picking methods use one feature to pick up the phase arrival,such as amplitude,energy,kurtosis.However,background and stationary noise often contaminate the seismic time-varing data,those methods perform well with low SNR.With high SNR,their results are unreliable.Motivated by these facts,we extract 15 dimensional features from time domain and frequency domain.The features are phase,amplitude,polarization,fractal dimension,STA/LTA,ML,amplitude ratio,energy ratio,curve length,kurtosis,skewness,RMS amplitude,average energy,amplitude kurtosis and frequency.2.By improving the local linear embedding algorithm,the data samples are reduced.Through the study of the time series,a method of kernel classification for time series is proposed.Combining the above methods,a P-wave automatic picking method(AMPAT)based on multidimensional feature is proposed.First,15-dimensional features from seismic signal are extracted as data set.Next,the improved LLE method is utilized to conduct the feature selection of the data set.And then,a new distance for time series and a simple inner product computing method are investigated,consequently,a kernel-based time series classification method is also presented.Finally,the proposed methods are used to determine the arrival time of P-wave.Experimental results on the real data sets demonstrate that the proposed method has promising performance over the comparedmethods.3.Many automatic phase picking methods are affected greatly by noise and their accuracy are depended on experienced threshold.So,an improved method DSLKPw based on DWT,STA/LTA and Kurtosis is proposed.The method firstly analyzes the seismic time-varing data via DWT to obtain the detail information and then analyzes the detail information using the STA/LTA method to reduce the influence of noise on P-wave picking.Finally the method implements the Kurtosis method to pick P-wave precisely,Simulation results show that the DSLKPw has the advantage of high accuracy of STA/LTA method and it also has the advantage of Kurtosis method which is not constrained by experienced threshold... |
PDF Full Text Request