地震层析成像LSQR算法的并行化

讨论了地震层析成像的LSQR算法(最小二乘QR分解). 在建立偏导数矩阵方程组时,对区内地震在方程中保留震源项,引入正交投影算子进行参数分离,对区外远震采用传统的平滑处理方式,用LSQR法求解联立的方程组. 由于区内地震的正交分解处理和区外远震的平滑处理,使得偏导数矩阵中的非零元素成倍增加,对于大型反演问题,这些非零元素常常达到几十GB到几百GB的数量级,巨量的内存占用成为LSQR算法的瓶颈. 针对这一问题,本文研究了偏导数矩阵中非零元素的分布规律,设计出合理的存储结构,采用分布式存储进行矩阵计算,提出了LSQR算法的并行化方案,并在联想深腾6800超级计算机上实现. 导出了LSQR算法的并行效率估算公式. 对两个地区的实际地震层析成像数据进行了效率测试.     

基于MPI的一维大地电磁并行计算研究

在大地电磁中运用并行处理技术来减少计算时间,提高运算效率.结合一维广义逆矩阵法反演的计算特点,详细分析了串行程序并行化方法,主要采用主从并行模式、分频并行计算的并行方案.在Linux系统上使用Fortran和MPICH2相结合的开发工具编写了并行程序,通过理论模型和实测数据对实现的并行程序进行试算,并与串行程序对比验证了该算法的可行性、正确性,为二维、三维的正反演并行计算提供了研究基础.     

重力梯度全张量数据三维共轭梯度聚焦反演

随着全张量重力梯度测量技术的日趋成熟和应用领域的不断扩大,重力梯度全张量数据的三维反演技术越来越受重视.本文利用剖分单元之间几何架构等效性,实现了重力梯度全张量场三维正演快速计算和导数矩阵优化存储.并将积分灵敏度、粗糙度和最小支撑泛函约束以及参考模型和模型参数界限约束引入到目标函数中,采用共轭梯度法进行反演迭代,实现了重力梯度全张量数据三维快速正反演计算.多种模型的反演试算表明:本文提出的反演算法的可靠性和稳定性较好,并且算法速度快、占用内存低且易于并行化.     

细胞自动机地震波模拟的并行化算法

采用区域分解的办法对格子气自动机地震波传播模拟的并行化算法进行了研究,提出了两种区域分解算法,并在SGIPowerChallenge并行计算机上进行了实际地质模型中平面波反射地震响应的计算,获得了较高的并行计算效率．     

模拟地震波传播的三维逐元并行谱元法

高效地震波场正演模拟对于复杂模型中地震波传播与成像研究至关重要.本文在谱元法原理框架内,对已有逐元谱元法改进,提出一种新的逐元并行谱元法求解三维地震波运动方程,并得到地震波场.逐元并行谱元法的核心思想在于在单元上进行质量矩阵与解向量的乘积运算,并将此运算平均分配至每一个CPU计算核心,此处理有利提升谱元法的并行计算效率.同时,根据Gauss-Lobatto-Legendre(GLL)数值积分点与插值点重合的特点,将稠密单元刚度矩阵的存储转化成单元雅克比矩阵行列式的值及其逆的存储,大幅减少谱元法计算内存开销.此外,在模型边界上利用逐元并行谱元法求解二阶位移形式完美匹配层(PML)吸收边界条件,消除边界截断而引入的虚假反射.通过逐元并行谱元法得到的数值解与解析解对比,以及实际地震波场模拟,数值结果证实了逐元并行谱元法用于地震波场模拟的高效性.     

基于改进的散射积分算法的初至波走时层析

初至波走时层析是获取近地表速度结构的一种常用方法.随着采集技术的不断发展,可使用的数据量迅速增多,传统的基于射线追踪和解方程组的地震走时层析成像方法面临着内存占用大、方程求解不稳定等问题.为了解决这些问题,本文基于前人在波形反演研究中提出的一种改进的散射积分算法,提出了一种预条件最速下降法初至波走时层析.该方法无需存储核函数矩阵与Hessian矩阵即可方便地实现目标函数梯度的计算与预条件,且该方法计算效率高、求解稳定、易于并行.数值实验结果表明,该方法可以获得与传统方法精度相当的反演结果,但所占用的内存大幅减小.     

基于紧凑存储的任意四边形网格有限元法地震波场数值模拟

采用有限元法精确模拟复杂介质条件下的地震波场,一般采用三角单元,但在节点数相同的情况下,三角单元的计算精度不如矩形单元高,采用三角单元模拟复杂界面时,编制确定结构刚度矩阵非零元素位置的程序也较为麻烦。采用矩形单元离散含有倾斜或起伏界面的地质模型时,无法避免绕射噪声,若加密网格又会增加计算量。为此,本文基于任意四边形单元模拟声波的传播,在倾斜或起伏界面条件下,可以有效避免因离散引起的“阶梯状”界面,在不增加计算量以及内存占用的前提下,有效地消除离散绕射噪声。采用对角的集中质量矩阵代替一致质量矩阵,避免矩阵的求逆运算,从而提高显式有限元法的计算效率;对结构刚度矩阵采用紧凑存储格式,每一行需要存储的元素最多为5个,同时零元素不参与运算,既减少内存的占用,又极大地提高计算效率。      

压缩存储约化法在水准网间接平差中的应用

本文提出了压缩存储约化法求解对称变带宽方程组的一种实用计算公式和程序编制。该方法的计算精度高于广义平方根分解法并探讨了压缩存储约化法在水准网间接平差中的应用。利用法方程系数阵的变带宽稀疏对称性,应用该法可使元素存入量由原来的n×n减少到5n—100n,大大节省了存储量和计算时间。 文中提出了作水准网的静态与动态秩亏网平差时,由程序自动提取法方程系数阵的压缩存储信息、用观测值信息直接累加式组成压缩存储的法方程系数阵及协因数阵的快速求出等具体计算方法。这些方法通用于任何水准网,可编通用程序。 文中列举了应用压缩存储约化法的效益,在PDP—11/23机(内存256KB)上双精度解算1366阶法方程的动态秩亏网平差,只占终端时间158分钟。而解算502阶法方程的静态平差,仅占终端时间9分钟。 本文最后给出了算例。     

用积分方程法模拟各向异性地层中三维电性异常体的电磁响应

本文研究并建立了一种模拟各向异性地层中三维电性异常体电磁响应的积分方程算法.首先讨论了并矢Green函数及其相关积分的计算,将水平层状各向异性地层中的电场并矢Green函数分解成含有奇异项的直达波与非奇异的来自各个层界面的反射和透射波两个部分,再应用等效体积单元和表面积分技术对积分方程的奇异核进行离散化处理以便提高离散方程的精度.然后为了节省计算机内存以及计算时间,引入基于Krylov子空间的迭代算法求解积分方程的离散化矩阵方程.最后通过与现有文献中的结果作对比从而检验了所述算法的有效性,并结合具体算例考察分析了地层的各向异性对三维电性异常体电磁响应的影响特征和规律.     

弹性波的三维有限元模拟

本文讨论了弹性波瞬态传播问题的三维有限元计算方法及当前存在的实际困难.针对要求计算机内存大和计算时间长的问题,采取了改进措施.由于采用了集中质量矩阵和修正的中心差分时间积分显格式相结合的方法,可以使计算机内存和计算时间大为减少;由于采用结点定位法,最适合用于目前发展的并行计算机系统,可使计算速度大大增快;还采用了有效激发震源法,有效激发区是随时间步进的增加而逐步增大,这不仅能节省计算时间,而且使波场的传播过程一目了然,本文计算了由两种介质组成的三维楔形问题,得到若干典型剖面的瞬时波场图及随时间变化的合成地震图.     

Modal time history analysis of non-classically damped structures for seismic motions

The step-by-step modal time history integration methods are developed for dynamic analysis of non-classically damped linear structures subjected to earthquake-induced ground motions. Both the mode displacement and mode acceleration-based algorithms are presented for the calculation of member and acceleration responses. The complex-valued eigenvectors are used to effect the modal decoupling of the equations of motion. However, the recursive step-by-step algorithms are still in terms of real quantities. The numerical results for the acceleration response and floor response spectra, obtained with these approaches, are presented. The mode acceleration approach is observed to be decidedly better than the mode displacement approach in as much as it alleviates the so-called missing mass effect, caused by the truncation of modes, very effectively. The utilization of the mode acceleration-based algorithms is, thus, recommended in all dynamic analyses for earthquake-induced ground motions.     

Diurnal periodicity of earthquakes in central Asia: Part 5. Catalogue of the Kyrgyz digital broadband seismic network KNET

Detailed analysis of the information contained in the earthquake catalogue of the Kyrgyz digital broadband seismic network over the period from 1994 to 2005 is performed. The graphs of earthquake recurrence were plotted and analyzed. The energy threshold of representativeness of the catalogue was determined. Stroboscopic and Fourier spectra of time series of earthquakes in the range of periods T ≤ 48 h are presented. A similarity between the obtained graph of diurnal course of seismicity with similar graphs for other regions was found. Peculiarities of diurnal course of earthquakes of different energy were studied. Earthquake sampling of the southwestern part of the territory presented in the catalogue was studied separately. Erroneous information in the catalogue caused by explosions is identified. A conclusion is made about a need for special measures to identify the explosions contained in the catalogue. Special algorithms for processing digital seismograms should be used in addition to collecting information about the time of explosions.     

一种线性走时插值射线追踪改进算法

针对线性走时插值算法(LTI)不能正确追踪逆向传播射线的问题, 目前已提出多种改进算法, 如扩张收缩LTI算法、 循环计算LTI算法、 动态网络最短路径射线追踪算法等, 但这些算法的计算效率普遍偏低. 在分析各种改进LTI算法的优劣后, 本文提出了改进动态网络最短路径射线追踪算法. 该改进算法依据波的传播规律以及LTI算法的基本方程, 排除动态网络最短路径射线追踪算法中大量冗余节点计算, 并采用传统的二叉树堆排序算法对波前阵列节点进行管理. 数值算例表明, 本文提出的改进算法具有较高的计算效率, 其计算效率是动态网络最短路径射线追踪算法的4.5—30倍, 是原始LTI算法的2—6.5倍; 当动态网络最短路径射线追踪算法采用堆排序算法时, 改进算法的计算效率是其3.5—15倍.      

Technologies of preliminary data processing for multidisciplinary geophysical monitoring and a case study of their application in the Kamchatka geoacoustic observation system

The problems and procedures for obtaining high quality results of multidisciplinary geophysical monitoring at the stage of preliminary data processing are considered. The case study of solving different technical problems during the long-term multidisciplinary borehole geophysical observations at the Petropavlovsk- Kamchatsky geodynamic research area is described. The algorithms for preliminary data processing and flawed data discard combining automated approaches and expert screening developed by the authors are presented. Suppression of quasi-regular disturbances is especially important in the studies of interrelations between geoacoustic, electromagnetic, and meteorological processes and seismicity because of the clear diurnal periodicity in all these processes that significantly complicates the signal extraction from the noise. The key technical, organizational, and methodical measures aimed at improving the data should be provided at the project planning stage of the measurement system. The optimal selection and reasoned application of special procedures for data preprocessing can crucially affect the research results.     

地震偏移成像不确定性分析初探

勘探地球物理领域存在大量宏观动力学过程的物理数学表达及尚待深入探索的不确定性事实,其中一类是源自对偶量的不可交换性或不可对易性.本文从地震偏移成像中的不确定性分析入手,探讨偏移成像中受到波场延拓过程不确定性的影响.文中给出了若干深度偏移算法对Marmousi模型的偏移不确定度分布表达及图像显示;文中还对若干波场延拓方法,从不确定性估计的角度进行了比较.笔者认为对应用技术领域中的宏观不确定性分析,将有益于深化对宏观动力学过程的理性认识,也为偏移算法评价提供一种途径;重要的是,进行不确定性分析将有利于把握所用偏移算法的不确定度分布,继而有针对性地采取相应的改善措施,以期提高算法的整体确定性程度.     

积分方程法复杂介质可控源电磁勘探快速正反演研究现状及发展方向

复杂介质可控源电磁勘探数值模拟及反演算法的研究一直是国内外地球物理学者研究的热点。本文对复杂介质可控源电磁勘探快速正反演算法研究进行综述,重点对复杂介质快速正反演算法及应用进行分析,指出高效并行、特殊边界条件或将是其真正实用化的关键,当前仍然是极具挑战的研究方向。着重对未受关注的可控源电磁法复杂介质积分方程法正反演算法及其应用研究,如二维、2.5维快速正反演算法;地面、井筒电磁勘探实例、起伏地形异常场模拟等进行讨论。指出国内积分方程法的研究相对滞后,但应用前景较可观;特别是大尺度隐伏资源勘探领域,高精度、高效电磁勘探正反演需求较迫切。通过体积分方程法快速正反演算例分析,表明该方法可适用于大尺度勘探生产,具有较好实用性。复杂地形模拟,高效正反演算法等是积分方程法实用化的关键。      

基于正交投影和广义投影(POCS)算法的研究进展

目的 在于提请有关读者关注近年来关于图像重建的凸集投影算法的进展，这一重要的方法及应用。方法 主要讨论了基于正交投影和广义投影的算法。对基于正交投影的算法，并讨论了一般形式的加权松驰格式，这包含了分块格式和同时格式。并以三个定理报道有关的算法收敛性结果。结果 分别包含了相容和不相容条件和弱强收敛下的结果。对基于广义投影的算法，有关的基本概念和例子，基本算法的收敛性结果。结论 报道最近关于引入松弛系数的工作和在CT图像重建中的应用。     

New algorithms of emission tomography for passive seismic monitoring of a producing hydrocarbon deposit: Part II. Results of real data processing

Two new algorithms of seismic emission tomography and the results of their numerical testing are presented. The algorithms allow eliminating of the screening effect caused by spatially correlated noise, and imaging of deep weak sources. Both algorithms invoke spatial filters, calculated on the basis of singular decomposition of the spectral matrix of the wave field. The key parameter of such a filter is the number of eigenvectors of the spectral matrix, which must be rejected in the process of seismic emission imaging in order to eliminate the screening effect. Consecutive increasing in the number of rejected eigenvectors enables imaging of ever weaker seismic sources.     

New algorithms of emission tomography for passive seismic monitoring of a producing hydrocarbon deposit: Part I. Algorithms of processing and numerical simulation

Two new algorithms of seismic emission tomography and the results of their numerical testing are presented. The algorithms allow eliminating of the screening effect caused by spatially correlated noise, and imaging of deep weak sources. Both algorithms invoke spatial filters, calculated on the basis of singular decomposition of the spectral matrix of the wave field. The key parameter of such a filter is the number of eigenvectors of the spectral matrix, which must be rejected in the process of seismic emission imaging in order to eliminate the screening effect. Consecutive increasing in the number of rejected eigenvectors enables imaging of ever weaker seismic sources.     

Collocation,dissipation and [overshoot] for time integration schemes in structural dynamics

The concept of collocation, originally used by Wilson in the development of dissipative algorithms for structural dynamics, is systematically generalized and analysed. Optimal schemes within this class are developed and compared with a recently proposed family of dissipative algorithms, called a methods. The α methods are found to be superior on the basis of standard measures of dissipation and dispersion. It is pointed out that the tendency to overshoot is an important and independent factor which should be considered in an evaluation of an implicit scheme. The basis for studying overshoot is discussed and the optimal collocation and α methods are compared. It is found that pathological overshooting is an inherent property of collocation schemes, whereas the overshooting characteristics of the α methods are good.