【作者】 赵亮；

【导师】 赵春霞；

【作者基本信息】 南京理工大学 ， 模式识别与人工智能， 2010， 博士

【摘要】 目前三维地震勘探已成为寻找油气资源的主要方法,而三维地震勘探数据是人工地震波对地下层位界面的反射信号。解释人员可以根据地下层位的起伏形态来推测地质结构,所以层位追踪是地震数据解释的重要基础工作。地下的层位在构造应力作用下破裂、错位,形成了断层。断层既是油气田的边界,也是油气运移、聚集的通道。因此断层的识别和建模是油气勘探开发中最重要的工作。目前断层识别是地震数据解释中最难的部分。识别断层的一种有效方法是三维相干技术,三维相干数据反映了原始地震数据的不连续属性,原始三维地震数据上不易发现的断层可以在三维地震相干切片上得到清晰的显示。本文对三维地震数据断层检测及建模的有关方法进行了研究,主要内容如下：1.现在常用的三维地震数据相干算法都是针对原始地震振幅数据的,计算量大,稳定性低。本文提出了一种基于梯度方向一致性的三维地震数据相干算法。该方法先估计体数据的方向场,然后根据体数据中各点的方向一致性计算地震数据的相干属性。该相干算法对数据局部区域的方向差异很敏感,可以作为断层检测的可靠依据。本文还进一步把该算法的结果应用到层位自动追踪及断层自动解释中。2.对于层位散点数据的曲面重构问题,本文提出了一种基于核回归的曲面拟合算法。该算法能较好地利用层位数据的特点,重构的计算量较小,重构曲面的面片数较少,还可以通过调整滤波参数兼顾曲面的平滑度和拟合精度。实验表明该方法的效率比传统的重构算法有较大提高,重构出的曲面无需后续平滑等处理,效果也能满足层位显示的需求。3.针对被复杂断层错裂的层位面重构问题,本文提出了一种重构破裂层位面的方法。该方法使用了迭代细分和能量逼近的方法,构造出一个逼近散点数据的连续曲面,再通过删除断层区域的网格面片来处理曲面上非连续的区域。该方法可以较好地重构含复杂断层的不连续层位面,较好的表现了层位的断裂特征。实验表明其重构精度较高,层位上的破裂部位显示效果较好。4.研究了由多层截面上的断层散点轮廓线重构三维断层面的问题,提出了一种基于最短连通路径的多层轮廓线拟合方法。该方法与人工拼接三维断层面的机理是相似的,即先在单独的一个截面上进行处理,通过二维平面上的散点构造最短路径轮廓,再进一步由多层轮廓线重构断层的曲面。实验表明该方法可以拟合出反映断层细节特征的光顺曲面。更多还原

【Abstract】 3D seismic survey is now the primary method in oil and gas exploration. The signal collected from the survey is the reflection of artificial seismic waves. The interpreters speculate the geological structures by the rolling shapes of the underground horizons, so the horizon picking is the foundation of seismic data interpretation. The fault identification is the difficulty in the seismic interpretation work. The fault is the breaking and misplacement of horizon underground, so it is the boundary of the oilfields and it is also the channel for oil to transport and gather. The fault identification is the most important work in oil exploration. The most effective method to detect faults is 3D coherence technique. The 3D coherence cube is the discontinuity attribute of the seismic data cube. The faults can not be found in the seismic data will display in the slices of coherence cube clearly.This thesis studies some problems about the 3D seismic data the fault detection and modeling. The main results obtained in the thesis are as follows:1. The current seismic coherence cube is for the seismic amplitude data, so the computational cost is high, and the stability is poor. This thesis proposed a coherence cube algorithm based on gradient orientation. This algorithm estimates the orientation field first, defines the coherence value by the local difference of the orientation vectors. It is sensitive to the local discontinuity of the seismic data. The result can be used to assist fault detection. This thesis uses this algorithm for horizon tracking and automatic interpretation of the faults.2. Attention goes to the reconstruction of scattered points from horizon picking. This thesis proposed a surface fitting algorithm based on kernel regression. This algorithm utilizes the characteristics of horizon points. The computational cost is low, and the result number of the reconstruction surface is small. The user can set the filter parameters to control smoothing degree and fitting precision. The experiments show that the method has far less complexity than traditional reconstruction algorithm, and the result surface can satisfy the horizon rendering requirement.3. The problem of reconstructing horizons which were torn by faults is addressed. This thesis proposed a iterative subdivision and approximation method to construct a continuous surface to fit the scattered points, then delete the fault area triangles. This method handles the broken horizons well, and the experiments show that the reconstruction result is precise and the rendering result of fault area is good.4. The fault surface reconstruction by multi-slice contours is studied. We proposed a algorithm based on the shortest contours on multi-slices. The method is like the process of artificial splicing, which processes a single slice first. The method gets the shortest path on every slice, constructs a fitting surface by multi-slice contours. The experiments show that the algorithm can get a smooth surface with the fault details.更多还原