A robust weighted total least squares algorithm and its geodetic applications

Total least squares (TLS) can solve the issue of parameter estimation in the errors-invariables (EIV) model, however, the estimated parameters are affected or even severely distorted when the observation vector and coefficient matrix are contaminated by gross errors. Currently, the use of existing robust TLS (RTLS) methods for the EIV model is unreasonable. Original residuals are directly used in most studies to construct the weight factor function, thus the robustness for the structure space is not considered. In this study, a robust weighted total least squares (RWTLS) algorithm for the partial EIV model is proposed based on Newton-Gauss method and the equivalent weight principle of general robust estimation. The algorithm utilizes the standardized residuals to construct the weight factor function and employs the median method to obtain a robust estimator of the variance component. Therefore, the algorithm possesses good robustness in both the observation and structure spaces. To obtain standardized residuals, we use the linearly approximate cofactor propagation law for deriving the expression of the cofactor matrix of WTLS residuals. The iterative procedure and precision assessment approach for RWTLS are presented. Finally, the robustness of RWTLS method is verified by two experiments involving line fitting and plane coordinate transformation. The results show that RWTLS algorithm possesses better robustness than the general robust estimation and the robust total least squares algorithm directly constructed with original residuals.     

Data-snooping procedure applied to errors-in-variables models

The theory of Baarda’s data snooping — normal and F tests respectively based on the known and unknown posteriori variance — is applied to detect blunders in errors-invariables (EIV) models, in which gross errors are in the vector of observations and/or in the coefficient matrix. This work is a follow-up to an earlier work in which we presented the formulation of the weighted total least squares (WTLS) based on the standard least squares theory. This method allows one to directly apply the existing body of knowledge of the least squares theory to the errors-in-variables models. Among those applications, data snooping methods in an EIV model are of particular interest, which is the subject of discussion in the present contribution. This paper generalizes the Baarda’s data snooping procedure of the standard least squares theory to an EIV model. Two empirical examples, a linear regression model and a 2-D affine transformation, using simulated and real data are presented to show the efficacy of the presented formulation. It is highlighted that the method presented is capable of detecting outlying equations (rather than outlying observations) in a straightforward manner. Further, the WTLS method can be used to handle different TLS problems. For example, the WTLS problem for the conditions and mixed models, the WTLS problem subject to constraints and variance component estimation for an EIV model can easily be established. These issues are in progress for future publications.     

A weighted least-squares solution to a 3-D symmetrical similarity transformation without linearization

A weighted least-squares (WLS) solution to a 3-D non-linear symmetrical similarity transformation within a Gauss-Helmert (GH) model, and/or an errors-in-variables (EIV) model is developed, which does not require linearization. The geodetic weight matrix is the inverse of the observation dispersion matrix (second-order moment). We suppose that the dispersion matrices are non-singular. This is in contrast to Procrustes algorithm within a Gauss-Markov (GM) model, or even its generalized algorithms within the GH and/or EIV models, which cannot accept geodetic weights. It is shown that the errors-invariables in the source system do not affect the estimation of the rotation matrix with arbitrary rotational angles and also the geodetic weights do not participate in the estimation of the rotation matrix. This results in a fundamental correction to the previous algorithm used for this problem since in that algorithm, the rotation matrix is calculated after the multiplication by row-wise weights. An empirical example and a simulation study give insight into the efficiency of the proposed procedure.     

A structured and constrained Total Least-Squares solution with cross-covariances

A new proof is presented of the desirable property of the weighted total least-squares (WTLS) approach in preserving the structure of the coefficient matrix in terms of the functional independent elements. The WTLS considers the full covariance matrix of observed quantities in the observation vector and in the coefficient matrix; possible correlation between entries in the observation vector and the coefficient matrix are also considered. The WTLS approach is then equipped with constraints in order to produce the constrained structured TLS (CSTLS) solution. The proposed approach considers the correlation between the observation vector and the coefficient matrix of an Error-In-Variables model, which is not considered in other, recently proposed approaches. A rigid transformation problem is done by preservation of the structure and satisfying the constraints simultaneously.     

Total least-squares adjustment of condition equations

The usual least-squares adjustment within an Errors-in-Variables (EIV) model is often described as Total Least-Squares Solution (TLSS), just as the usual least-squares adjustment within a Random Effects Model (REM) has become popular under the name of Least-Squares Collocation (without trend). In comparison to the standard Gauss-Markov Model (GMM), the EIV-Model is less informative whereas the REM is more informative. It is known under which conditions exactly the GMM or the REM can be equivalently replaced by a model of condition equations or, more generally, by a Gauss-Helmert Model. Similar equivalency conditions are, however, still unknown for the EIV-Model once it is transformed into such a model of condition equations. In a first step, it is shown in this contribution how the respective residual vector and residual matrix look like if the TLSS is applied to condition equations with a random coefficient matrix to describe the transformation of the random error vector. The results are demonstrated using a numeric example which shows that this approach may be valuable in its own right.     

A general weighted total Kalman filter algorithm with numerical evaluation

An applicable algorithm for Total Kalman Filter (TKF) approach is proposed. Meanwhile, we extend it to the case in which we can consider arbitrary weight matrixes for the observation vector, the random design matrix and possible correlation between them. Also the updated dispersion matrix of the predicted unknown is given. This approach makes use of condition equations and straightforward variance propagation rules. It is applicable to data fusion within a dynamic errors-in-variables (DEIV) model, which usually appears in the determination of the position and attitude of mobile sensors. Then, we apply for the first time the TKF algorithm and its extended version named WTKF to a DEIV model and compare the results. The results show the efficiency of the proposed WTKF algorithm. In particular in the case of large weights, WTKF shows approximately 25% improvement in contrast to TKF approach.     

Variance component estimation in errors-in-variables models and a rigorous total least-squares approach

A method for variance component estimation (VCE) in errors-in-variables (EIV) models is proposed, which leads to a novel rigorous total least-squares (TLS) approach. To achieve a realistic estimation of parameters, knowledge about the stochastic model, in addition to the functional model, is required. For an EIV model, the existing TLS techniques either do not consider the stochastic model at all or assume approximate models such as those with only one variance component. In contrast to such TLS techniques, the proposed method considers an unknown structure for the stochastic model in the adjustment of an EIV model. It simultaneously predicts the stochastic model and estimates the unknown parameters of the functional model. Moreover the method shows how an EIV model can support the Gauss-Helmert model in some cases. To make the VCE theory into EIV model more applicable, two simplified algorithms are also proposed. The proposed methods can be applied to linear regression and datum transformation. We apply these methods to these examples. In particular a 3-D non-linear close to identical similarity transformation is performed. Two simulation studies besides an experimental example give insight into the efficiency of the algorithms.     

Look-ahead vertical seismic profiling inversion approach for density and compressional wave velocity in Bayesian framework

To reduce drilling uncertainties, zero-offset vertical seismic profiles can be inverted to quantify acoustic properties ahead of the bit. In this work, we propose an approach to invert vertical seismic profile corridor stacks in Bayesian framework for look-ahead prediction. The implemented approach helps to successfully predict density and compressional wave velocity using prior knowledge from drilled interval. Hence, this information can be used to monitor reservoir depth as well as quantifying high-pressure zones, which enables taking the correct decision during drilling. The inversion algorithm uses Gauss–Newton as an optimization tool, which requires the calculation of the sensitivity matrix of trace samples with respect to model parameters. Gauss–Newton has quadratic rate of convergence, which can speed up the inversion process. Moreover, geo-statistical analysis has been used to efficiently utilize prior information supplied to the inversion process. The algorithm has been tested on synthetic and field cases. For the field case, a zero-offset vertical seismic profile data taken from an offshore well were used as input to the inversion algorithm. Well logs acquired after drilling the prediction section was used to validate the inversion results. The results from the synthetic case applications were encouraging to accurately predict compressional wave velocity and density from just a constant prior model. The field case application shows the strength of our proposed approach in inverting vertical seismic profile data to obtain density and compressional wave velocity ahead of a bit with reasonable accuracy. Unlike the commonly used vertical seismic profile inversion approach for acoustic impedance using simple error to represent the prior covariance matrix, this work shows the importance of inverting for both density and compressional wave velocity using geo-statistical knowledge of density and compressional wave velocity from the drilled section to quantify the prior covariance matrix required during Bayesian inversion.     

Unscented transformation with scaled symmetric sampling strategy for precision estimation of total least squares

The errors-in-variables (EIV) model is a nonlinear model, the parameters of which can be solved by singular value decomposition (SVD) method or the general iterative algorithm. The existing formulae for covariance matrix of total least squares (TLS) parameter estimates don’t fully consider the randomness of quantities in iterative algorithm and the biases of parameter estimates and residuals. In order to reflect more reasonable precision information for TLS adjustment, the derivative-free unscented transformation with scaled symmetric sampling strategy, i.e. scaled unscented transformation (SUT), is introduced and implemented. In this contribution, we firstly discuss the existing various solutions of TLS adjustment and covariance matrices of TLS parameter estimates and derive the general first-order approximate cofactor matrices of random quantities in TLS adjustment. Secondly, based on the combination of TLS iterative algorithm and calculation process of SUT, we design the two SUT algorithms to calculate the biases and the second-order approximate covariance matrices. Finally, the straight line fitting model and plane coordinate transformation model are used to demonstrate that applying SUT for precision estimation of TLS adjustment is feasible and effective.     

基于PCA-PSO-SVM的地震死亡人数预测模型研究

为准确预测地震死亡人数,提出了基于主成分分析法(PCA)和粒子群算法(PSO)优化的支持向量机(SVM)模型。首先利用主成分分析法对地震死亡人数7个影响因子中的6个进行数据降维,同时对第7个发震时刻因子单独进行区间分类,然后对提取出的主成分进行归一化处理,将归一化的主成分数据作为支持向量机的输入向量,通过粒子群算法寻优获得最优支持向量机模型参数,最终建立基于PCA-PSO-SVM的地震死亡人数预测模型,并对5组样本进行死亡人数预测,同时对比分析包含和不包含发震时刻因子的2种情况下的模型预测效果。结果表明:在不考虑发震时刻因子的情况下,使用PCA-PSO-SVM模型的最小误差、最大误差和平均误差分别为0.85%、20%、10%,其平均误差相比PSO-SVM、SVM模型分别降低2.08%、2.28%;输入向量加入发震时刻因子分类数据后,PCA-PSO-SVM模型的最小误差、最大误差和平均误差分别为0.25%、20%、7.18%,其平均误差相比PSO-SVM、SVM模型分别降低3.34%、3.50%。因此,加入发震时刻因子后3种模型的平均误差明显降低,同时由于PCA-PSO-SVM模型进行主成分降维处理,能够明显提高运行效率和预测精度,故降低了模型复杂度。     

Increasing numerical efficiency of iterative solution for total least-squares in datum transformations

Cartesian coordinate transformation between two erroneous coordinate systems is considered within the Errors-In-Variables (EIV) model. The adjustment of this model is usually called the total Least-Squares (LS). There are many iterative algorithms given in geodetic literature for this adjustment. They give equivalent results for the same example and for the same user-defined convergence error tolerance. However, their convergence speed and stability are affected adversely if the coefficient matrix of the normal equations in the iterative solution is ill-conditioned. The well-known numerical techniques, such as regularization, shifting-scaling of the variables in the model, etc., for fixing this problem are not applied easily to the complicated equations of these algorithms. The EIV model for coordinate transformations can be considered as the nonlinear Gauss-Helmert (GH) model. The (weighted) standard LS adjustment of the iteratively linearized GH model yields the (weighted) total LS solution. It is uncomplicated to use the above-mentioned numerical techniques in this LS adjustment procedure. In this contribution, it is shown how properly diminished coordinate systems can be used in the iterative solution of this adjustment. Although its equations are mainly studied herein for 3D similarity transformation with differential rotations, they can be derived for other kinds of coordinate transformations as shown in the study. The convergence properties of the algorithms established based on the LS adjustment of the GH model are studied considering numerical examples. These examples show that using the diminished coordinates for both systems increases the numerical efficiency of the iterative solution for total LS in geodetic datum transformation: the corresponding algorithm working with the diminished coordinates converges much faster with an error of at least 10^-5 times smaller than the one working with the original coordinates.     

基于Remez迭代算法求解差分系数的双相介质自适应有限差分正演模拟

利用传统有限差分方法对基于Biot理论的双相介质波动方程进行数值求解时,由于慢纵波的存在,数值频散效应较为明显,影响模拟精度.相对于声学近似方程及普通弹性波方程,Biot双相介质波动方程在同等数值求解算法和精度要求条件下,其地震波场正演模拟需要更多的计算时间.本文针对Biot一阶速度-应力方程组发展了一种变阶数优化有限差分数值模拟方法,旨在同时提高其正演模拟的精度和效率.首先结合交错网格差分格式推导Biot方程的数值频散关系式.然后基于Remez迭代算法求取一阶空间偏导数的优化差分系数,并用于Biot方程的交错网格有限差分数值模拟.在此基础上把三类波的平均频散误差参数限制在给定的频散误差阈值和频率范围内,此时优化有限差分算子的长度就能自适应非均匀双相介质模型中的不同速度区间.数值频散曲线分析表明:基于Remez迭代算法的优化有限差分方法相较传统泰勒级数展开方法在大波数范围对频散误差的压制效果更明显;可变阶数的优化有限差分方法能取得与固定阶数优化有限差分方法相近的模拟精度.在均匀介质和河道模型的数值模拟实验中将本文变阶数优化有限差分算法与传统泰勒展开算法、最小二乘优化算法进行比较,进一步证明其在复杂地下介质中的有效性和适用性.     

基于BTTB矩阵的快速高精度三维磁场正演

本文改进了一种快速、高精度空间域三维正演算法,用来计算地下场源在水平观测面产生的磁异常ΔT场及其梯度场,以解决传统空间域正演计算效率低的问题.算法采用长方体对场源区域进行剖分,观测点与场源剖分单元体中心点在水平面上的投影重合.改进的算法具有以下三个特点:(1)采用无解析奇点的解析解公式计算磁异常,保证计算精度.(2)通...     

Variance-covariance component estimation for structured errors-in-variables models with cross-covariances

Studia Geophysica et Geodaetica - In this contribution, an iterative algorithm for variance-covariance component estimation based on the structured errors-in-variables (EIV) model is proposed. We...     

基于PCA-GSM-SVM的地震伤亡人数预测

针对影响地震伤亡人数的评价指标数量较多且各指标之间存在着复杂的非线性关系,运用机器学习理论,提出了基于支持向量机(Support Vector Machine)的地震伤亡人数预测模型;首先利用主成分分析法(Principle Component Analysis)对7个地震死亡人数影响指标进行数据降维,然后对提取出的主成分进行归一化处理,将归一化的主成分数据作为预测模型的输入向量,将地震伤亡人数作为预测模型的输出向量;以27个地震伤亡实例作为学习样本进行训练,运用网格搜索法(Grid Search Method)寻优获得最优支持向量机参数,最终建立基于PCA-GSM-SVM的地震死亡人数预测模型,并对5组样本进行死亡人数预测。结果表明:PCA-GSM-SVM模型的最小误差、最大误差和平均误差分别为5.12%、15.7%和9.16%,其平均误差相比于GSM-SVM模型和SVM模型分别降低6.51%和7.11%,因此PCA-GSM-SVM模型预测精度较高,可在工程实际中推广。     

三维陆地可控源电磁法有限元模型降阶快速正演

三维陆地可控源电磁法有限元快速正演的主要瓶颈在于多频率大型稀疏方程组求解问题.本文引入一种基于模型降阶的Krylov子空间投影算法,推导了有限元刚度矩阵的模型降阶形式,构建了频率域传递函数;采用标准正交向量序列,构建一个远远小于有限元刚度矩阵维度的矩阵,该矩阵与频率无关,通过一次模型降阶即可实现多频点有限元方程快速求解.采用基于电场的变分方程,加入散度校正条件,以消除伪解;引入伪δ函数,消除了源点的奇异性,可适用于复杂背景模型三维有限元数值模拟,并为多源的求解奠定了基础;以层状介质模型解析解为标准,通过和基于Pardiso直接求解器的有限元算法（3DFEM）进行比较,模型降阶法计算时间小于前者的1/10,平均相对误差在1.72%,在满足精度要求下,实现了高效率三维有限元数值求解;分别设计了横向高低阻模型和纵向高低阻模型,分析了从近区到远区电场和卡尼亚视电阻率的变化规律,假极值的表现特征,阴影效应的影响等,从而也验证了该算法的正确性.最后,建立了一个地层陷落柱模型,通过模型降阶有限元正演模拟,发现视电阻率断面图在陷落柱上方出现"凹陷",与模型设计吻合,表明该算法对复杂地层模拟具有同样的适用性.     

3D parallel inversion of time-domain airborne EM data

To improve the inversion accuracy of time-domain airborne electromagnetic data, we propose a parallel 3D inversion algorithm for airborne EM data based on the direct Gauss–Newton optimization. Forward modeling is performed in the frequency domain based on the scattered secondary electrical field. Then, the inverse Fourier transform and convolution of the transmitting waveform are used to calculate the EM responses and the sensitivity matrix in the time domain for arbitrary transmitting waves. To optimize the computational time and memory requirements, we use the EM “footprint” concept to reduce the model size and obtain the sparse sensitivity matrix. To improve the 3D inversion, we use the OpenMP library and parallel computing. We test the proposed 3D parallel inversion code using two synthetic datasets and a field dataset. The time-domain airborne EM inversion results suggest that the proposed algorithm is effective, efficient, and practical.     

A least-squares penalty method algorithm for inverse problems of steady-state aquifer models

Based on the generalized Gauss–Newton method, a new algorithm to minimize the objective function of the penalty method in (Bentley LR. Adv Wat Res 1993;14:137–48) for inverse problems of steady-state aquifer models is proposed. Through detailed analysis of the “built-in” but irregular weighting effects of the coefficient matrix on the residuals on the discrete governing equations, a so-called scaling matrix is introduced to improve the great irregular weighting effects of these residuals adaptively in every Gauss–Newton iteration. Numerical results demonstrate that if the scaling matrix equals the identity matrix (i.e., the irregular weighting effects of the coefficient matrix are not balanced), our algorithm does not perform well, e.g., the computation cost is higher than that of the traditional method, and what is worse is the calculations fail to converge for some initial values of the unknown parameters. This poor situation takes a favourable turn dramatically if the scaling matrix is slightly improved and a simple preconditioning technique is adopted: For naturally chosen simple diagonal forms of the scaling matrix and the preconditioner, the method performs well and gives accurate results with low computational cost just like the traditional methods, and improvements are obtained on: (1) widening the range of the initial values of the unknown parameters within which the minimizing iterations can converge, (2) reducing the computational cost in every Gauss–Newton iteration, (3) improving the irregular weighting effects of the coefficient matrix of the discrete governing equations. Consequently, the example inverse problem in Bentley (loc. cit.) is solved with the same accuracy, less computational effort and without the regularization term containing prior information on the unknown parameters. Moreover, numerical example shows that this method can solve the inverse problem of the quasilinear Boussinesq equation almost as fast as the linear one.In every Gauss–Newton iteration of our algorithm, one needs to solve a linear least-squares system about the corrections of both the parameters and the groundwater heads on all the discrete nodes only once. In comparison, every Gauss–Newton iteration of the traditional method has to solve the discrete governing equations as many times as one plus the number of unknown parameters or head observation wells (Yeh WW-G. Wat Resour Res 1986;22:95–108).All these facts demonstrate the potential of the algorithm to solve inverse problems of more complicated non-linear aquifer models naturally and quickly on the basis of finding suitable forms of the scaling matrix and the preconditioner.     

多频可控源电磁法三维有理函数Krylov子空间模型降阶正演算法研究

本文采用有理函数Krylov子空间模型降阶算法实现了同时求解多频可控源电磁法三维正演响应的快速计算.首先采用基于Yee氏交错网格的拟态有限体积法实现控制方程的空间离散,将任意频率的电场响应表示为关于频率参数的传递函数.采用有理函数Krylov子空间算法求解该传递函数.针对构建m维有理函数Krylov子空间需要求解m次（几十到上百）关于有理函数极点和离散控制方程系数矩阵的线性方程组的问题,本文提出采用单个重复极点的有理函数Krylov子空间模型降阶算法,结合直接法求解器PARDISO,采用Gram-Schmidt方法,只需要1次系数矩阵分解和m次矩阵回代即可实现有理函数Krylov子空间的构建,极大地减少了计算量.针对最优化有理函数极点选取问题,本文根据传递函数的有理函数Krylov子空间投影算法的误差分析理论,引入关于单个重复极点的收敛率函数,通过求解有理函数的最大收敛率直接给出最优化的单个重复极点公式.最终实现了不同发射频率的可控源电磁法三维正演响应的快速计算.分别计算了典型层状模型多发射频率的CSAMT和海洋CSEM的正演响应,通过与解析解的对比验证了本文算法在多发射频率正演的计算精度和计算效率;并通过一个三维海洋CSEM勘探设计最优化发射频率和接收区域选取的例子进一步说明本文算法的优点.     

有限元三角网格中波动的频散分析及数值仿真

波动问题有限元离散后会引起数值误差, 数值频散的本质就是数值误差传播引起的非物理解. 数值频散不仅没有实际意义, 而且还会影响对真实波动现象的认识. 为厘清有限元三角网格中波动数值频散的影响因素, 本文推导了集中质量矩阵和一致质量矩阵的频散函数, 同时给出了组合质量矩阵的频散函数, 并对不同质量矩阵的数值频散进行了对比研究. 理论分析和数值计算结果表明: 有限元三角网格中波动的数值频散受网格布局、 波传播方向、 单元网格纵横比以及质量矩阵的影响; 一致质量矩阵的数值频散比集中质量矩阵更易受到波传播方向的影响; 不合理的三角网格单元会对数值相速度(数值频散)产生不良影响; 正三角网格中波动的数值频散几乎不受波传播方向的影响; 一致质量矩阵与集中质量矩阵的线性组合能够有效地压制数值频散.