Research On Seismic Horizon Tracking Method Based On Signal Classification

With the rapid development of modern information technology, data incommunication, biology and other fields appear explosive growth, which makes thesignal classification become more and more import. The signal classification is aimedto determine the type of signal by extracting and analyzing the characteristicparameters of the signal. This paper commences the work with two-dimensional andthree-dimensional seismic images as research object, signal classification as a researchmethod, extracting horizon from seismic images as research purpose.Horizon tracking plays a crucial role in the analysis and interpretation of theseismic data when predicting the reservoir of oil, natural gas. Horizon extraction is amethod designed to extracting the lineups of the reflection layer on the geologic profilefrom the seismic image obtained by the exploration. In this thesis, we make an in-depthstudy on the general process of signal classification, solve the special problems andsituations faced by the seismic images, and develop a method of horizon trackingapplied to actual data. Specific work as follows:1. For the point that the existing all-horizon extraction methods can’t achieve anaccurate extraction of all the horizons within the designated area, this thesis proposed amethod of all-horizon tracking based on the signal classification. Firstly, this methodreconstructs the seismic waveform by polynomial fitting, the polynomial fitcoefficients as the parameters of the waveform characteristics; Secondly, the methoduses FSSCEM (Feature Subset Selection Component-wise EM) algorithm to optimizeand cluster the set of obtained characteristic parameters to obtain the optimized set ofcharacteristic parameters and the corresponding results of optimized clustering; Finallywe make use of the direction of the inclination to search by correlation, fill the horizon,remove the overlapping horizon and combine all the segments to achieve the extractionof all the horizons within the designated area in2D seismic data.2. For the problem of horizon tracking in3-D seismic data, this thesis proposes anew extraction method of horizon surface in three-dimensional seismic images in thisthesis. Firstly, this method converts the tracking in3D data into the tracking in the two-dimensional profiles according to the information of the seed points. Secondly,convert the problem of horizon tracking in three-dimensional seismic images into theproblem of semi-supervised classification, using the seed points to identify the targethorizon. Finally this method gets the complete surface of three-dimensional layers byfilling the horizon using searching by correlation with the direction of inclination andother subsequent processing.The proposed algorithm is applied to the data processing of a work area inSichuan. From the point of the result on processing actual work area, the proposedmethod in this thesis has achieved good result. The method proposed in this thesis canaccurately extract all horizons in the specified area and obtain better results than thecommercial software OpendTect in terms of the all-horizon tracking in2-D image ofseismic profile. The extraction method of the horizon surface in the three-dimensionalimage proposed in this thesis can extract the horizon surface in3-D seismic imageaccurately and obtain a better result than manual tracking in terms of resolution interms of the extracting of the surface of horizon in3-D seismic image.