鲁棒性自适应共轭梯度子波估计方法

发布时间：2018-04-22 03:32

本文选题：地震子波估计 + 共轭梯度　；参考：《西南交通大学》2015年硕士论文

【摘要】：在地震勘探工作中,对地震资料进行反褶积、波阻抗反演、AVO反演以及正演模型的建立这些工作都依赖于高精度的地震子波。而在实际勘探过程中,地震子波常常是未知的,必须通过我们已有的地震资料提取出真实的子波,常规的子波提取方法有直接法、确定性方法和统计性方法：直接法就是利用探测仪器直接提取地震子波,确定性估计方法就是在测井资料已知的前提下,首先由测井曲线计算出地层反射系数,然后根据地震褶积模型结合地震数据提取出真实子波,其最大的优点就是可直接计算出真实子波而不需要对子波和反射系数进行假设,但是对地震资料井震匹配程度,测井数据的准确性依赖比较大；统计性子波估计方法是仅通过地震道自身的二阶或高阶统计特性来提取子波,该方法不需要测井数据,但需要对地层反射系数和子波做出一定的假设。本文主要针对测井处和井旁地震道进行研究。在测井数据已知的情况下,利用测井声波和密度资料计算出实际地层反射系数进行确定性子波估计。通过对地震褶积模型的观察,我们可以将地震记录的形成过程理解为,地层反射系数信号经过子波系统滤波后得到的输出信号即为地震道记录,并将褶积模型下的子波估计问题理解为自适应信号处理中的系统识别问题,从这个角度出发利用自适应滤波算法来估计出最优的子波。在自适应算法中,LMS类算法具有结构简单容易实现、计算复杂度低等优点,其最大的缺点就是收敛慢,而RLS类算法虽然能快速收敛,但其计算复杂,需要较大的存储空间进行矩阵运算,且存在数值不稳定。共轭梯度(CG)算法是一种在性能上介于LMS和RLS之间的算法,它在收敛快的同时具有较低的计算复杂度。因此,本文选择共轭梯度滤波算法来研究地震子波估计问题。通过对现有子波估计方法进行研究,针对子波估计的实际问题,提出相应的解决方案。针对在实际地震勘探中子波长度的不确定性,利用高阶累计量MA模型阶数判定方法粗略的估计出地震子波长度；考虑到大多数地震数据长度有限,提出利用递归块方法来提高共轭梯度算法的收敛性能,使得算法能在有限次迭代内收敛；传统的子波估计方法往往都是在背景噪声为白噪声或色噪声假设下进行的,考虑到地震勘探环境复杂,既有可能存在脉冲噪声的情形,结合M-估计算法提出针对非高斯噪声的鲁棒性子波估计方法；观察通过对实际子波波形的观察,我们发现子波能量比较集中,且存在较长的零区间,这里我们就可以将子波理解为稀疏或半稀疏信号,通过对目标函数加入估计子波稀疏性约束提高算法性能和子波的精确度。理论模型仿真和实际地震资料处理都表明,本文提出的算法能有效的抑制高斯和脉冲非高斯噪声,快速有效的提取出精确的子波。
[Abstract]:In seismic exploration, seismic data deconvolution, wave impedance inversion and AVO inversion, as well as the establishment of forward model, all depend on high-precision seismic wavelet. In the actual exploration process, the seismic wavelet is often unknown. We must extract the real wavelet from the existing seismic data. The conventional wavelet extraction method has the direct method. Deterministic method and statistical method: the direct method is to extract seismic wavelet directly by means of detecting instruments. The deterministic estimation method is to calculate the formation reflection coefficient from the log curve on the premise that the log data is known. Then the real wavelet is extracted according to the seismic convolution model and seismic data. Its biggest advantage is that the real wavelet can be directly calculated without the assumption of wavelet and reflection coefficient, but the well earthquake matching degree of seismic data can be obtained. The accuracy of logging data depends heavily on the statistical wavelet estimation method, which only uses the second-order or high-order statistical characteristics of the seismic trace itself to extract the wavelet, which does not require the logging data. But it is necessary to make some assumptions about the formation reflection coefficient and wavelet. This paper mainly focuses on the well logging and the seismic trace around the well. Under the condition that the log data are known, the reflection coefficient of the actual formation is calculated by using the log acoustic wave and density data to estimate the deterministic wavelet. By observing the seismic convolution model, we can interpret the formation process of seismic records as that the output signals of formation reflection coefficient signals filtered by wavelet system are seismic trace records. The wavelet estimation problem based on convolution model is interpreted as a system recognition problem in adaptive signal processing. From this point of view, the optimal wavelet estimation is obtained by using adaptive filtering algorithm. In the adaptive algorithm, the LMS-class algorithm has the advantages of simple structure, low computational complexity and so on. Its biggest disadvantage is the slow convergence, while the RLS class algorithm can converge quickly, but its calculation is complex. Large storage space is needed for matrix operation, and the numerical value is unstable. The conjugate gradient CGG algorithm is an algorithm between LMS and RLS in performance. It has fast convergence and low computational complexity. Therefore, the conjugate gradient filtering algorithm is chosen to study the problem of seismic wavelet estimation. By studying the existing wavelet estimation methods, a corresponding solution is proposed to solve the practical problems of wavelet estimation. In view of the uncertainty of neutron wave length in actual seismic exploration, seismic wavelet length is roughly estimated by using the order determination method of high order cumulant MA model, considering that the length of most seismic data is limited. A recursive block method is proposed to improve the convergence performance of the conjugate gradient algorithm, so that the algorithm can converge within a finite iteration, and the traditional wavelet estimation methods are usually carried out under the assumption that the background noise is white or colored noise. Considering the complexity of seismic exploration environment and the possibility of impulse noise, a robust wavelet estimation method for non- noise is proposed in combination with the M- estimation algorithm, and the actual wavelet waveform is observed. We find that wavelet energy is concentrated and there is a long zero interval. Here we can interpret wavelet as sparse or semi-sparse signal and improve the algorithm performance and wavelet accuracy by adding wavelet sparsity constraints to the objective function. Theoretical model simulation and actual seismic data processing show that the proposed algorithm can effectively suppress Gao Si and impulse non- noise and extract accurate wavelet quickly and effectively.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：P631.4

【参考文献】