Linearized seismic waveform inversion using a multiple re-weighted least-squares method with QR preconditioning

Linearized inversion methods such as Gauss‐Newton and multiple re‐weighted least‐squares are iterative processes in which an update in the current model is computed as a function of data misfit and the gradient of data with respect to model parameters. The main advantage of those methods is their ability to refine the model parameters although they have a high computational cost for seismic inversion. In the Gauss‐Newton method a system of equations, corresponding to the sensitivity matrix, is solved in the least‐squares sense at each iteration, while in the multiple re‐weighted least‐squares method many systems are solved using the same sensitivity matrix. The sensitivity matrix arising from these methods is usually not sparse, thus limiting the use of standard preconditioners in the solution of the linearized systems. For reduction of the computational cost of the linearized inversion methods, we propose the use of preconditioners based on a partial orthogonalization of the columns of the sensitivity matrix. The new approach collapses a band of co‐diagonals of the normal equations matrix into the main diagonal, being equivalent to computing the least‐squares solution starting from a partial solution of the linear system. The preconditioning is driven by a bandwidth L which can be interpreted as the distance for which the correlation between model parameters is relevant. To illustrate the benefit of the proposed approach to the reduction of the computational cost of the inversion we apply the multiple re‐weighted least‐squares method to the 2D acoustic seismic waveform inversion problem. We verify the reduction in the number of iterations in the conjugate'gradient algorithm as the bandwidth of the preconditioners increases. This effect reduces the total computational cost of inversion as well.     

Three‐term inversion of prestack seismic data using a weighted l2, 1 mixed norm

We present a new inversion method to estimate, from prestack seismic data, blocky P‐ and S‐wave velocity and density images and the associated sparse reflectivity levels. The method uses the three‐term Aki and Richards approximation to linearise the seismic inversion problem. To this end, we adopt a weighted mixed l_{2, 1}‐norm that promotes structured forms of sparsity, thus leading to blocky solutions in time. In addition, our algorithm incorporates a covariance or scale matrix to simultaneously constrain P‐ and S‐wave velocities and density. This a priori information is obtained by nearby well‐log data. We also include a term containing a low‐frequency background model. The l_{2, 1} mixed norm leads to a convex objective function that can be minimised using proximal algorithms. In particular, we use the fast iterative shrinkage‐thresholding algorithm. A key advantage of this algorithm is that it only requires matrix–vector multiplications and no direct matrix inversion. The latter makes our algorithm numerically stable, easy to apply, and economical in terms of computational cost. Tests on synthetic and field data show that the proposed method, contrarily to conventional l₂‐ or l₁‐norm regularised solutions, is able to provide consistent blocky and/or sparse estimators of P‐ and S‐wave velocities and density from a noisy and limited number of observations.     

快速精确的二维频率空间域弹性波数值模拟

二维频率空间域的数值模拟方法具有以下的优势:多炮模拟时,计算成本比时间域方法低;无累计误差;在地震反演中处理多震源模拟时,只需要有限的几个频率就可以得到好的反演结果.差分离散化形成的稀疏系数矩阵,需要求解一个巨大规模的线性方程组,最大瓶颈是需要海量的计算机内存,导致计算量庞大.本文在前人研究的基础上,采用嵌套剖分网格排序法,极大限度减少对计算机内存的需求,从而减少了计算量.针对弹性波数值模拟的特征,提出二维频率空间域弹性波多炮模拟的快速计算流程.数值模拟试验证明使用嵌套剖分排序法的弹性波多炮数值模拟比压缩存储法具有节省存储量、计算效率高等优势,为后续的二维频率空间域弹性波全波形反演奠定了很好的基础.     

基于高斯牛顿法的频率域可控源电磁三维反演研究

三维反演解释是电磁法勘探发展的重要趋势,而如何提高三维反演的可靠性、稳定性和计算效率是算法开发者们目前的研究重点.本文实现了一种频率域可控源电磁(CSEM)三维反演算法.其中正演基于拟态有限体积法离散化,利用直接矩阵分解技术来求解大型线性系统方程,不仅准确、稳定,而且特别有利于含有大量发射场源位置的CSEM勘探情况;对目标函数的最优化采用高斯牛顿法(GN),具有近似二次的收敛性;使用预条件共轭梯度法(PCG)求解每次GN迭代所得到的法方程,避免了显式求解和存储灵敏度矩阵,减小了计算量.以上这些方法的结合应用,使得本文的三维反演算法准确、稳定且高效.通过陆地和海洋CSEM勘探场景中的典型理论模型的反演测试,验证了本文算法的有效性.     

Tidal flow forecasting using reduced rank square root filters

The Kalman filter algorithm can be used for many data assimilation problems. For large systems, that arise from discretizing partial differential equations, the standard algorithm has huge computational and storage requirements. This makes direct use infeasible for many applications. In addition numerical difficulties may arise if due to finite precision computations or approximations of the error covariance the requirement that the error covariance should be positive semi-definite is violated. In this paper an approximation to the Kalman filter algorithm is suggested that solves these problems for many applications. The algorithm is based on a reduced rank approximation of the error covariance using a square root factorization. The use of the factorization ensures that the error covariance matrix remains positive semi-definite at all times, while the smaller rank reduces the number of computations and storage requirements. The number of computations and storage required depend on the problem at hand, but will typically be orders of magnitude smaller than for the full Kalman filter without significant loss of accuracy. The algorithm is applied to a model based on a linearized version of the two-dimensional shallow water equations for the prediction of tides and storm surges. For non-linear models the reduced rank square root algorithm can be extended in a similar way as the extended Kalman filter approach. Moreover, by introducing a finite difference approximation to the Reduced Rank Square Root algorithm it is possible to prevent the use of a tangent linear model for the propagation of the error covariance, which poses a large implementational effort in case an extended kalman filter is used.     

基于静主元消元法的频率域波动方程正演(英文)

频率域波动方程正演是求解一个大型线性稀疏方程组问题,其受到计算效率和内存存储问题的限制。常规的高斯消元法不能满足大型数据的并行计算,本文提出基于静主元消元法(GESP)进行稀疏矩阵LU分解和多炮有限差分正演,该方法不仅提高了稳定性,更有利于单频点内LU分解的分布式并行计算。通过Marmousi模型模拟试验,单频波场和转化到时间域地震剖面的试验表明模拟精度和计算效率得到提高,节约并充分利用内存,为波形反演奠定基础。     

Visco-acoustic transmission waveform inversion of velocity structure in space-frequency domain

According to the least square criterion of minimizing the misfit between modeled and observed data,this paper provides a preconditioned gradient method to invert the visco-acoustic velocity structure on the basis of using sparse matrix LU factorization technique to directly solve the visco-acoustic wave forward problem in space-frequency domain.Numerical results obtained in an inclusion model inversion and a layered homogeneous model inversion demonstrate that different scale media have their own frequency ...     

频率域粘弹性声波透射波形速度反演

在用稀疏矩阵的LU分解技术对频率域粘弹性声波方程进行直接求解的基础上,根据失配函数二范数最小准则, 用预条件梯度类方法对粘弹性声波介质的速度结构进行了逐频反演. 局部非均匀介质模型和层状介质模型速度结构反演的实验结果表明,不同频率能够反映地下介质的多尺度物性结构（低频数据对应与介质物性的大尺度结构）,用低频反演结果作为高频反演的初值逼近这一顺序模式,能大大改善反演过程中解的非唯一性. 而且,在反演过程中用Hess矩阵的对角线元素来做梯度类方法的预条件算子,能够吸收了高斯牛顿法的二次收敛优势, 使得本文算法具有较快的收敛速度.      

基于不等式约束的最小二乘法三维电阻率反演及其算法优化

基于光滑约束的最小二乘法是三维电阻率反演的主要方法,但该方法在某些情况下存在着多解性较强的问题,且普遍耗时较长,严重制约了三维反演方法的推广与发展.为改善上述问题,将表征模型参数变化范围的不等式约束作为先验信息引入最小二乘线性反演方法中,有效地改善了反演结果的精度,降低了反演的多解性问题.为了解决耗时较长的问题,基于预条件共轭梯度(PCG)算法和Cholesky分解法的特点提出了一套优化三维电阻率反演计算效率的计算方案.在该方案中,Cholesky分解法被用来求解敏感度矩阵计算中的多个点源场的正演问题,Cholesky分解法只需对总体系数矩阵进行一次分解,然后对不同的右端向量进行回代即可.将预条件共轭梯度法引入到三维电阻率反演方程的求解中,将雅可比迭代中的对角阵作为预处理矩阵,其具有求逆方便、无需内存空间的特点,有效地加快了收敛速度.对合成数据以及实测数据的反演算例表明,借助不等式约束和反演效率优化方案,最小二乘反演方法可得到较为精确的反演结果,有效地提高了反演计算效率,具有良好的推广前景.     

反射地震走时层析成像中的大型稀疏矩阵压缩存储和求解

反射地震走时层析成像是一种精度较高的速度求取方法,最终可归结为线性方程组的求解.方程组具有很大的维数,常规解法需要很大的存储量和计算量.本文考虑到当投影函数取为走时残差,图象函数取为慢度残差时,灵敏度矩阵中的元素表示射线经过网格的长度的特殊物理意义,采用行索引的压缩存储方式,在射线追踪正演模拟过程中直接压缩存储灵敏度矩阵,在层析反演过程中利用压缩后的矩阵进行求解,大大降低了存储量和计算量.     

Two‐dimensional waveform inversion of multi‐component data in acoustic‐elastic coupled media

In order to account for the effects of elastic wave propagation in marine seismic data, we develop a waveform inversion algorithm for acoustic‐elastic media based on a frequency‐domain finite‐element modelling technique. In our algorithm we minimize residuals using the conjugate gradient method, which back‐propagates the errors using reverse time migration without directly computing the partial derivative wavefields. Unlike a purely acoustic or purely elastic inversion algorithm, the Green's function matrix for our acoustic‐elastic algorithm is asymmetric. We are nonetheless able to achieve computational efficiency using modern numerical methods. Numerical examples show that our coupled inversion algorithm produces better velocity models than a purely acoustic inversion algorithm in a wide variety of cases, including both single‐ and multi‐component data and low‐cut filtered data. We also show that our algorithm performs at least equally well on real field data gathered in the Korean continental shelf.     

Using surface multiples to estimate primaries by sparse inversion from blended data

For data acquired with conventional acquisition techniques, surface multiples are usually considered as noise events that obscure the primaries. However, in this paper we demonstrate that for the situation of blended acquisition, meaning that different sources are shooting in a time‐overlapping fashion, multiples can be used to ‘deblend’ the seismic measurements. We utilize the recently introduced estimation of primaries by sparse inversion (EPSI) methodology, in which the primary impulse responses are considered to be the unknowns in a large‐scale inversion process. With some modifications the estimation of primaries by sparse inversion method can be used for blended seismic data. As output this process gives unblended primary impulse responses with point sources and receivers at the surface, which can be used directly in traditional imaging schemes. It turns out that extra information is needed to improve on the deblending of events that do not have much associated multiple energy in the data, such as steep events at large offsets. We demonstrate that this information can be brought in during acquisition and during processing. The methodology is illustrated on 2D synthetic data.     

2.5D forward modeling and inversion of frequency-domain airborne electromagnetic data

Frequency-domain airborne electromagnetics is a proven geophysical exploration method. Presently, the interpretation is mainly based on resistivity—depth imaging and one-dimensional layered inversion; nevertheless, it is difficult to obtain satisfactory results for two- or three-dimensional complex earth structures using 1D methods. 3D forward modeling and inversion can be used but are hampered by computational limitations because of the large number of data. Thus, we developed a 2.5D frequency-domain airborne electromagnetic forward modeling and inversion algorithm. To eliminate the source singularities in the numerical simulations, we split the fields into primary and secondary fields. The primary fields are calculated using homogeneous or layered models with analytical solutions, and the secondary (scattered) fields are solved by the finite-element method. The linear system of equations is solved by using the large-scale sparse matrix parallel direct solver, which greatly improves the computational efficiency. The inversion algorithm was based on damping least-squares and singular value decomposition and combined the pseudo forward modeling and reciprocity principle to compute the Jacobian matrix. Synthetic and field data were used to test the effectiveness of the proposed method.     

Preconditioned non‐linear conjugate gradient method for frequency domain full‐waveform seismic inversion

We present preconditioned non‐linear conjugate gradient algorithms as alternatives to the Gauss‐Newton method for frequency domain full‐waveform seismic inversion. We designed two preconditioning operators. For the first preconditioner, we introduce the inverse of an approximate sparse Hessian matrix. The approximate Hessian matrix, which is highly sparse, is constructed by judiciously truncating the Gauss‐Newton Hessian matrix based on examining the auto‐correlation and cross‐correlation of the Jacobian matrix. As the second preconditioner, we employ the approximation of the inverse of the Gauss‐Newton Hessian matrix. This preconditioner is constructed by terminating the iteration process of the conjugate gradient least‐squares method, which is used for inverting the Hessian matrix before it converges. In our preconditioned non‐linear conjugate gradient algorithms, the step‐length along the search direction, which is a crucial factor for the convergence, is carefully chosen to maximize the reduction of the cost function after each iteration. The numerical simulation results show that by including a very limited number of non‐zero elements in the approximate Hessian, the first preconditioned non‐linear conjugate gradient algorithm is able to yield comparable inversion results to the Gauss‐Newton method while maintaining the efficiency of the un‐preconditioned non‐linear conjugate gradient method. The only extra cost is the computation of the inverse of the approximate sparse Hessian matrix, which is less expensive than the computation of a forward simulation of one source at one frequency of operation. The second preconditioned non‐linear conjugate gradient algorithm also significantly saves the computational expense in comparison with the Gauss‐Newton method while maintaining the Gauss‐Newton reconstruction quality. However, this second preconditioned non‐linear conjugate gradient algorithm is more expensive than the first one.     

稀疏测线大地电磁资料三维反演研究:合成算例(英文)

受勘探成本和工区环境等因素的影响,当前大多数大地电磁实际工作采取布置稀疏测线采集数据和使用二维反演方法解释这些稀疏测线数据的方式。然而,二维反演方法在解释三维地电构造数据时存在局限性,有时甚至做出错误的地质解释。本文尝试了使用三维反演方法对大地电磁稀疏测线数据进行反演解释。使用大地电磁全信息资料三维共轭梯度反演程序对理论模型合成稀疏测线数据进行了三维反演。结果表明:这种反演方案是可行与有效的。同时,我们发现在不同数据的三维反演结果中,四个张量阻抗元素和两个倾子数据同时反演的结果相对更为准确,更接近理论模型。     

Accelerating the T‐matrix approach to seismic full‐waveform inversion by domain decomposition

A seismic variant of the distorted Born iterative inversion method, which is commonly used in electromagnetic and acoustic (medical) imaging, has been recently developed on the basis of the T‐matrix approach of multiple scattering theory. The distorted Born iterative method is consistent with the Gauss–Newton method, but its implementation is different, and there are potentially significant computational advantages of using the T‐matrix approach in this context. It has been shown that the computational cost associated with the updating of the background medium Green functions after each iteration can be reduced via the use of various linearisation or quasi‐linearisation techniques. However, these techniques for reducing the computational cost may not work well in the presence of strong contrasts. To deal with this, we have now developed a domain decomposition method, which allows one to decompose the seismic velocity model into an arbitrary number of heterogeneous domains that can be treated separately and in parallel. The new domain decomposition method is based on the concept of a scattering‐path matrix, which is well known in solid‐state physics. If the seismic model consists of different domains that are well separated (e.g., different reservoirs within a sedimentary basin), then the scattering‐path matrix formulation can be used to derive approximations that are sufficiently accurate but far more speedy and much less memory demanding because they ignore the interaction between different domains. However, we show here that one can also use the scattering‐path matrix formulation to calculate the overall T‐matrix for a large model exactly without any approximations at a computational cost that is significantly smaller than the cost associated with an exact formal matrix inversion solution. This is because we have derived exact analytical results for the special case of two interacting domains and combined them with Strassen's formulas for fast recursive matrix inversion. To illustrate the fact that we have accelerated the T‐matrix approach to full‐waveform inversion by domain decomposition, we perform a series of numerical experiments based on synthetic data associated with a complex salt model and a simpler two‐dimensional model that can be naturally decomposed into separate upper and lower domains. If the domain decomposition method is combined with an additional layer of multi‐scale regularisation (based on spatial smoothing of the sensitivity matrix and the data residual vector along the receiver line) beyond standard sequential frequency inversion, then one apparently can also obtain stable inversion results in the absence of ultra‐low frequencies and reduced computation times.     

频率域海洋可控源电磁垂直各向异性三维反演

地层宏观电性各向异性会对可控源电磁响应产生重要影响.由于海底地层电性结构常表现为电导率各向异性,若仅对海洋可控源电磁(MCSEM)数据进行常规各向同性反演,有可能无法获得准确的反演解释结果,从而削弱MCSEM技术的可靠性.本文实现了电导率垂直各向异性(VTI)条件下频率域海洋可控源电磁数据三维反演算法.其中,三维正演采用基于二次场控制方程的交错网格有限体积法,并利用直接矩阵分解技术来求解离散所得的大型线性方程组,有利于快速计算多场源的响应.反演采用具有近似二次收敛性的高斯牛顿算法对目标函数进行最优化.最后,对具有VTI电性各向异性特征的盐丘构造模型的MCSEM合成数据分别进行了电导率各向同性和垂直各向异性三维反演,结果表明:各向同性三维反演算法无法对受VTI介质影响的MCSEM数据进行正确的反演解释,而垂直各向异性三维反演能够获得更为可靠的地下电阻率结构和异常体分布,展现出对海底电性各向异性结构更为优良的反演解释能力.     

Misfit functionals in Laplace‐Fourier domain waveform inversion,with application to wide‐angle ocean bottom seismograph data

In seismic waveform inversion, non‐linearity and non‐uniqueness require appropriate strategies. We formulate four types of L2 normed misfit functionals for Laplace‐Fourier domain waveform inversion: i) subtraction of complex‐valued observed data from complex‐valued predicted data (the ‘conventional phase‐amplitude’ residual), ii) a ‘conventional phase‐only’ residual in which amplitude variations are normalized, iii) a ‘logarithmic phase‐amplitude’ residual and finally iv) a ‘logarithmic phase‐only’ residual in which the only imaginary part of the logarithmic residual is used. We evaluate these misfit functionals by using a wide‐angle field Ocean Bottom Seismograph (OBS) data set with a maximum offset of 55 km. The conventional phase‐amplitude approach is restricted in illumination and delineates only shallow velocity structures. In contrast, the other three misfit functionals retrieve detailed velocity structures with clear lithological boundaries down to the deeper part of the model. We also test the performance of additional phase‐amplitude inversions starting from the logarithmic phase‐only inversion result. The resulting velocity updates are prominent only in the high‐wavenumber components, sharpening the lithological boundaries. We argue that the discrepancies in the behaviours of the misfit functionals are primarily caused by the sensitivities of the model gradient to strong amplitude variations in the data. As the observed data amplitudes are dominated by the near‐offset traces, the conventional phase‐amplitude inversion primarily updates the shallow structures as a result. In contrast, the other three misfit functionals eliminate the strong dependence on amplitude variation naturally and enhance the depth of illumination. We further suggest that the phase‐only inversions are sufficient to obtain robust and reliable velocity structures and the amplitude information is of secondary importance in constraining subsurface velocity models.     

基于密度模型稀疏表征的重力反演方法

重力反演是恢复地下密度空间分布的有效工具,而选择合理的密度模型约束方法是提升重力反演分辨率和可靠性的关键.常规约束方法大多是从剖分网格空间中的密度模型出发,通过调整光滑或稀疏约束权重来匹配反演目标,但当地质体类型多样、异常分离不准确及网格剖分方案不合理时,模型约束的合理性与灵活性难以得到有效保证.为此,本文提出了一种基...     

Gravity compression forward modeling and multiscale inversion based on wavelet transform

The main problems in three-dimensional gravity inversion are the non-uniqueness of the solutions and the high computational cost of large data sets. To minimize the high computational cost, we propose a new sorting method to reduce fluctuations and the high frequency of the sensitivity matrix prior to applying the wavelet transform. Consequently, the sparsity and compression ratio of the sensitivity matrix are improved as well as the accuracy of the forward modeling. Furthermore, memory storage requirements are reduced and the forward modeling is accelerated compared with uncompressed forward modeling. The forward modeling results suggest that the compression ratio of the sensitivity matrix can be more than 300. Furthermore, multiscale inversion based on the wavelet transform is applied to gravity inversion. By decomposing the gravity inversion into subproblems of different scales, the non-uniqueness and stability of the gravity inversion are improved as multiscale data are considered. Finally, we applied conventional focusing inversion and multiscale inversion on simulated and measured data to demonstrate the effectiveness of the proposed gravity inversion method.