Seismic Signal Feature Extraction And Application Based On EEMD And SVM

Seismic signal is a non-stationary and non-linear signal formed by seismic wave after earth filtering.Seismic attributes contain a large amount of formation information,which can accurately reflect the formation morphology and characteristics.With these attributes,reservoir fluids can be effectively identified.Time-frequency analysis is an important part of reservoir identification and interpretation technology.With this method,the time-frequency spectrum of seismic signals can be obtained,and then the characteristics of the time-frequency domain are used to extract seismic attributes and identify reservoir fluids.The traditional time-frequency analysis method is essentially a type of Fourier transform based on linear stationary signals and adjustable windows,and has low time-frequency resolution and cross-term problems,which limits its application in seismic exploration.Hilbert-Huang Transform(HHT)is an adaptive time-frequency analysis method based on the characteristics of the signal itself.It can greatly improve the time-frequency resolution of the seismic signal.However,this method has problems of modal aliasing,endpoint effects,and instantaneous attribute errors,which will cause great interference to subsequent interpretation work.This paper mainly improves the shortcomings of the HHT method,and then based on high-precision time-frequency analysis,extracts seismic attributes from post-stack seismic data in the western Sichuan region,and uses Support Vector Machine(SVM)to realize adaptive identification of reservoir fluid.The main contents of this paper are as follows:(1)Contrastive study of various traditional time-frequency analysis methods.For example,Short-Time Fourier Transform,Wavelet Transform,Wigner-Ville Transform,S transform,etc.Based on this,the HHT method is introduced to process the seismic signals.The principle of the method,the Empirical Mode Decomposition(EMD)process and the implementation process of Hilbert transform to extract the instantaneous attributes are systematically analyzed.Through comparative analysis,the advantages of HHT method in processing non-stationary signals with higher time-frequency resolution are highlighted.(2)Perform a comprehensive analysis of the defects in the conventional HHT and improve it to form a complete set of improved HHT methods.Among them,the Ensemble Empirical Mode Decomposition(EEMD)method that makes the signal continuous at different time scales by adding Gaussian white noise is used to effectively improve the modal aliasing;the Auto-regressive(AR)model is used to reduce the endpoint effect on EEMD Adverse effects caused by decomposition;use the Normalized Hilbert Transform(NHT)to suppress the problem of transient attribute errors.(3)The EEMD method is used to extract the seismic attributes of the actual seismic signals in western Sichuan and optimize them.First,the Intrinsic Mode Function(IMF)component obtained from the EEMD decomposition is optimized by combining the correlation coefficient and the log profile,so that it highlights the reflective layer information of the reservoir fluid.Then,The NHT method was then used to extract 10 seismic attributes closely related to the reservoir fluid.Finally,the attribute fusion technology is used to optimize the extracted attributes to avoid the multiple solutions of a single attribute and improve the accuracy and efficiency of reservoir identification.(5)SVM model has strong advantages in solving nonlinear,small sample and high-dimensional pattern recognition.Train the SVM recognition model by using the optimized four attributes as SVM input feature quantities.Using evaluation indexes such as accuracy,recall and time,the performance of the grid search method,genetic algorithm and particle swarm optimization to optimize the parameter model is compared.At the same time,the validity of SVM model is verified by comparison with BP neural network.The analysis of data in western Sichuan proves that the SVM model optimized by parameters can effectively identify weak gas-bearing reservoirs.