【摘要】：隨著人工智能的發(fā)展,特征分類技術(shù)已經(jīng)成為一種重要的數(shù)據(jù)分析方式,被廣泛應(yīng)用在數(shù)據(jù)挖掘、人工智能、模式識別等領(lǐng)域。特征分類是通過分析和辨識數(shù)據(jù)的特征和類別的關(guān)系,產(chǎn)生一個(gè)關(guān)于特征和類別的模型,然后利用這個(gè)模型來分析和處理新的數(shù)據(jù)特征,確定新數(shù)據(jù)的類別。本文通過研究特征分類技術(shù)和三維地震數(shù)據(jù)中層位特征,設(shè)計(jì)了三維地震數(shù)據(jù)中的全層位追蹤方法。三維地震圖像是地震波形在三維空間的分布,層位追蹤是就是分析三維地震圖像中的層位信息。目前的層位追蹤工作主要依靠地質(zhì)解釋人員在二維剖面上進(jìn)行層位標(biāo)定,由于二維剖面并不能反映三維數(shù)據(jù)的全貌,人工解釋在反映地質(zhì)結(jié)構(gòu)全貌上和效率上存在嚴(yán)重的不足,嚴(yán)格意義上的三維全層位追蹤算法直接處理三維地震圖像,在準(zhǔn)確性和效率上存在很大的優(yōu)勢。本文就是針對上述問題對三維地震圖像中的層位追蹤問題進(jìn)行研究,主要工作包括以下兩個(gè)方面:(1)本文研究了三維地震圖像中的層位特征和數(shù)據(jù)挖掘中特征分類技術(shù),并結(jié)合三維地震圖像中的波形特征、層位極值點(diǎn)空間分布特征以及特征分類技術(shù)的一般方法設(shè)計(jì)了基于多層次框架的三維全層位自動追蹤方法,該方法將層位追蹤過程按照所處理數(shù)據(jù)的粒度劃分成了極值點(diǎn)層次、層位片段層次以及專家層次,分別從不同的層次研究三維地震圖像中的層位追蹤問題,在極值點(diǎn)層次根據(jù)極值點(diǎn)的空間上橫向連續(xù)分布的特征設(shè)計(jì)了基于層位極值點(diǎn)連接的層位片段生成算法,將極值點(diǎn)連接成層位片段;在層位片段層次根據(jù)同一層位的波形相似性,結(jié)合GMM聚類算法設(shè)計(jì)了層位片段合并算法,將層位片段連接成大的層位;在專家層次,為了使地質(zhì)解釋人員可以根據(jù)實(shí)際需求調(diào)整層位追蹤的結(jié)果,設(shè)計(jì)了專家級的層位結(jié)果修正算法,使得層位追蹤結(jié)果更加符合實(shí)際的地質(zhì)構(gòu)造。該方法在不同層次對層位追蹤問題進(jìn)行相應(yīng)的處理,最終實(shí)現(xiàn)了三維地震圖像中三維全自動追蹤。(2)本文從一種全新的角度提取三維地震圖像中層位特征,通過研究三維地震圖像中層位縱向分布特征提出了基于匹配搜索的全層位追蹤方法。傳統(tǒng)的層位追蹤方法都是逐個(gè)層位的進(jìn)行追蹤,這樣就忽視了三維地震圖像中目標(biāo)區(qū)域內(nèi)時(shí)間方向上不同層位之間的關(guān)系,在三維地震圖像中時(shí)間方向上層位呈現(xiàn)層狀分布,相鄰層位之間存在一定的間隙,而且不同層位之間的間隙會有一定的差異,這就是三維地震圖像中的層位縱向分布特征�；谄ヅ渌阉鞯娜珜游蛔粉櫡椒ㄖ饕鶕�(jù)時(shí)間方向上多個(gè)層位縱向分布的振幅特征和層位之間間隙的特征,設(shè)計(jì)了層位縱向分布特征提取算法、基于匹配搜索的數(shù)據(jù)塊生成算法和基于振幅導(dǎo)向的數(shù)據(jù)塊連接算法。該方法利用層位縱向分布特征提取算法和基于匹配搜索的數(shù)據(jù)塊生成算法將三維地震圖像中的極值點(diǎn)連接成三維空間中的層位塊,基于振幅導(dǎo)向的數(shù)據(jù)塊連接算法將層位塊進(jìn)行連接,形成地震圖像中大的層位,然后針對層位的缺口問題對層位進(jìn)行擴(kuò)展處理得到最終的追蹤結(jié)果。由于該方法充分利用了層位間的相互關(guān)系,實(shí)現(xiàn)了三維地震圖像中全層位并行追蹤,提高了層位追蹤的效率和準(zhǔn)確性。
[Abstract]:With the development of artificial intelligence, feature classification technology has become an important way of data analysis, and has been widely used in data mining, artificial intelligence, pattern recognition and other fields. This paper designs a full-layer tracking method in 3D seismic data by studying feature classification technology and horizon features in 3D seismic data. 3D seismic image is the distribution of seismic waveform in 3D space, horizon tracking is the analysis of horizon in 3D seismic image. The current work of horizon tracing mainly relies on geological interpreters to calibrate horizons on two-dimensional profiles. Because the two-dimensional profiles can not reflect the overall picture of three-dimensional data, there are serious shortcomings in the efficiency and efficiency of artificial interpretation in reflecting the overall picture of geological structure. Strict three-dimensional full-horizon tracing algorithm directly deals with three-dimensional data. Seismic images have great advantages in accuracy and efficiency. In this paper, the problem of horizon tracking in 3D seismic images is studied. The main work includes the following two aspects: (1) In this paper, horizon features in 3D seismic images and feature classification technology in data mining are studied and combined with 3D seismograms. Waveform features in images, spatial distribution characteristics of horizon extremum points and general method of feature classification technology are designed based on multi-level framework. This method divides the horizon tracking process into extremum point level, horizon segment level and expert level according to the granularity of the data processed, respectively, never. The problem of horizon tracking in 3-D seismic images is studied at the same level. An algorithm of horizon segment generation based on horizon extreme point connection is designed at the extreme point level according to the spatial continuous distribution characteristics of the extreme points. Clustering algorithm designed a layer fragment merging algorithm to connect the layer fragments into large layers; in the expert level, in order to make geological interpreters adjust the results of layer tracking according to actual needs, an expert level correction algorithm was designed to make the results of layer tracking more in line with the actual geological structure. The problem of horizon tracing is dealt with at the same level, and three-dimensional automatic tracing is realized in the end. (2) In this paper, horizon features in three-dimensional seismic images are extracted from a new perspective, and a full-horizon tracing method based on matching search is proposed by studying the vertical distribution characteristics of horizon in three-dimensional seismic images. The method of horizon tracing is to track each horizon one by one, thus ignoring the relationship between different horizons in the time direction of the target area in the three-dimensional seismic image. The horizons in the time direction of the three-dimensional seismic image show layered distribution, and there are certain gaps between adjacent horizons, and the gaps between different horizons will be one. According to the amplitude characteristics of the vertical distribution of multiple horizons in the time direction and the characteristics of the interval between horizons, the method of full horizon tracing based on matching search is designed to extract the vertical distribution of horizons, and the data block generation algorithm based on matching search is designed. In this method, the extremum points in 3-D seismic images are connected to the horizon blocks in 3-D space by using the vertical distribution feature extraction algorithm and the matching search-based data block generation algorithm, and the horizon blocks are connected by the amplitude-oriented data block connection algorithm to form the seismogram. The method makes full use of the interrelation between horizons, realizes the parallel tracking of all horizons in 3D seismic images, and improves the efficiency and accuracy of horizon tracking.
【學(xué)位授予單位】：電子科技大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2015
【分類號】：P631.4;TP391.41

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 李雪峰;閻建國;趙州;姚爽;;利用相干屬性剖面特征進(jìn)行層位解釋[J];物探化探計(jì)算技術(shù);2011年02期

，

本文編號：2198008