加权总体最小二乘的效率优化算法

加权总体最小二乘法是理论上估计EIV模型参数相对严密的方法,其迭代过程中涉及的矩阵运算较为耗时,在处理大量级数据时尤其明显。PEIV模型有助于提高加权总体最小二乘法的计算效率。本文基于PEIV模型和经典最小二乘准则给出了一种加权总体最小二乘法算法,算法的推导过程简洁,易于理解,迭代过程中无需重构矩阵,减少了矩阵运算量。最后通过仿真试验验证了算法的可靠性。试验结果表明,本文算法可以取得与现有算法相同的参数估计精度且计算效率更高。     

三维坐标转换的通用整体最小二乘算法

三维坐标转换模型属于非线性EIV（errors-in-variables）模型,现有整体最小二乘算法均设定了某些特殊假设条件,如仅适用于小角度或者属于非统计意义上的数值解,并且不能用于结构性的系数矩阵等,算法适用性受到极大限制。本文提出了三维坐标转换模型的通用加权整体最小二乘算法,该算法适用于任意旋转角度以及一般性的权矩阵情况下的三维坐标转换模型,并且将结构性系数矩阵、同时包含随机和非随机元素的系数矩阵等情况纳入到了统一的坐标转换模型算法。实例计算表明,本文提出的算法具有通用性,适用于实际应用中的各类三维坐标转换模型。     

一种相关观测的Partial EIV模型求解方法

Partial errors-in-variables(Partial-EIV)模型作为EIV模型的扩展形式,其构造方式更有规律,解算方法更为简便,能有效应用于实际情况。针对已有Partial EIV模型方法未考虑观测向量和系数矩阵存在相关性这一情况,通过提取观测向量和系数矩阵组成的增广矩阵中非重复出现的随机元素,构建更具一般适用性的Partial EIV模型,在该模型的基础上,将特殊假定条件扩展到不限定观测数据相关性的一般情况,详细推导了观测向量和系数矩阵元素相关且不等精度情况下的加权总体最小二乘方法,通过算例试验,并与目前已有的解决EIV模型相关观测情况下的方法进行了比较分析,研究表明本文方法可以提高计算效率,更具一般性,特别是对于观测向量和系数矩阵中存在常数元素和重复元素的情况。     

水准测量中尺度比参数的附加系统参数的Partial EIV模型解法

针对目前水准测量中尺度比参数的附加系统参数的平差模型,本文对模型进行重构,提出尺度比参数的附加系统参数的partial errors-in-variables（Partial EIV）模型,给出总体最小二乘准则下的解算公式及迭代算法。实际算例和模拟数据结果表明,本文方法与传统方法在处理尺度比参数为附加系统参数时的效果基本一致,并给出相关的理论依据。     

Partial EIV模型的解法

提出了一种求解partial errors-in-variables(partial EIV)模型的思路。通过对partial EIV模型的部分元素进行移项,重组成新形式下的平差函数模型,两次运用间接平差原理分别求解平差参数与系数矩阵中的随机元素,把总体最小二乘平差问题转化为最小二乘平差问题,并通过适当变换提高了新解法的收敛速度。最后分别采用实测数据和模拟数据进行验证,求解了本文算法与已有算法的估值结果。算例结果表明,本文算法能取得与已有算法相同的结果,是切实可行的。     

Statistical Methods for Jointly Estimating the Decay Constant of 40K and the Age of a Dating Standard

To simultaneously evaluate the decay constant of ⁴⁰K () and the age of a standard (t _std) using isotopic data from geologic materials, we applied a series of statistical methods. The problem of estimating the most probable intercept of many nonlinear curves in and t _std space is formulated by an errors-in-variables nonlinear regression model. Then a maximum likelihood method is applied to the model for a point estimate, which is equivalent to the nonlinear least square method when measurement error distributions are Gaussian. Uncertainties and confidence regions of the estimates can be approximated using three methods: the asymptotic normal approximation, the parametric bootstrap method and Bonferroni confidence regions. Five pairs of published data for samples with ages from 2 ka to 4.5 Ga were used to estimate and the age of Fish Canyon sanidine (t _FCs). The statistical procedure yields most probable estimates of (5.4755 ± 0.0170 × 10^–10 (1)/year) and t _FCs (28.269 ± 0.0661 (1) Ma) which are in between previously published values. These results indicate the power of our approach to provide improved constraints on these parameters, although the preliminary nature of some of the input data require further review before the values can be adopted.     

线性EIV模型的t型估计

将数理统计中的t型估计作为一种抗差估计引入测量平差中,提出线性EIV模型的t型抗差估计及其EM算法。以平面拟合为例,当向观测向量和系数矩阵加入粗差时,选取合适的自由度v,t型估计具有较好的抗粗差能力,求出的参数均值与真值偏差较小。     

附不等式约束的总体最小二乘迭代算法

基于惩罚函数和测量平差中权的思想,提出了附不等式约束的总体最小二乘平差模型,即利用惩罚函数对不等式约束方程构造约束权,通过零权和无限权将不等式约束转换为等式约束,从而将不等式约束平差准则转化为传统的测量平差准则。同时,根据非线性最小二乘平差理论,用构造结构矩阵的方法来顾及系数矩阵的结构性,推导了附不等式约束的总体最小二乘迭代算法。该算法迭代格式与传统的间接平差类似,只需经过若干次迭代便能得到最优解。     

A History of Regression and Related Model-Fitting in the Earth Sciences (1636?-2000)

The (statistical) modeling of the behavior of a dependent variate as a function of one or more predictors provides examples of model-fitting which span the development of the earth sciences from the 17th Century to the present. The historical development of these methods and their subsequent application is reviewed. Bond's predictions (c. 1636 and 1668) of change in the magnetic declination at London may be the earliest attempt to fit such models to geophysical data. Following publication of Newton's theory of gravitation in 1726, analysis of data on the length of a 1° meridian arc, and the length of a pendulum beating seconds, as a function of sin²(latitude), was used to determine the ellipticity of the oblate spheroid defining the Figure of the Earth. The pioneering computational methods of Mayer in 1750, Boscovich in 1755, and Lambert in 1765, and the subsequent independent discoveries of the principle of least squares by Gauss in 1799, Legendre in 1805, and Adrain in 1808, and its later substantiation on the basis of probability theory by Gauss in 1809 were all applied to the analysis of such geodetic and geophysical data. Notable later applications include: the geomagnetic survey of Ireland by Lloyd, Sabine, and Ross in 1836, Gauss's model of the terrestrial magnetic field in 1838, and Airy's 1845 analysis of the residuals from a fit to pendulum lengths, from which he recognized the anomalous character of measurements of gravitational force which had been made on islands. In the early 20th Century applications to geological topics proliferated, but the computational burden effectively held back applications of multivariate analysis. Following World War II, the arrival of digital computers in universities in the 1950s facilitated computation, and fitting linear or polynomial models as a function of geographic coordinates, trend surface analysis, became popular during the 1950–60s. The inception of geostatistics in France at this time by Matheron had its roots in meeting the evident need for improved estimators in spatial interpolation. Technical advances in regression analysis during the 1970s embraced the development of regression diagnostics and consequent attention to outliers; the recognition of problems caused by correlated predictors, and the subsequent introduction of ridge regression to overcome them; and techniques for fitting errors-in-variables and mixture models. Improvements in computational power have enabled ever more computer-intensive methods to be applied. These include algorithms which are robust in the presence of outliers, for example Rousseeuw's 1984 Least Median Squares; nonparametric smoothing methods, such as kernel-functions, splines and Cleveland's 1979 LOcally WEighted Scatterplot Smoother (LOWESS); and the Classification and Regression Tree (CART) technique of Breiman and others in 1984. Despite a continuing improvement in the rate of technology-transfer from the statistical to the earth-science community, despite an abrupt drop to a time-lag of about 10 years following the introduction of digital computers, these more recent developments are only just beginning to penetrate beyond the research community of earth scientists. Examples of applications to problem-solving in the earth sciences are given.     

带有变量中误差非正态模型中的Box-Cox变换

Box-Cox变换主要用于得到近似正态分布的随机变量。线性模型中的响应(因)变量Y(或经过变换的Y)通常建立在精确测量无误差的独立变量集上。本文研究独立变量中的测量误差(即变量中误差)对Box-Cox变换中参数估计的影响。研究结果表明,忽略变量中的误差将产生变换参数估计的偏离,该结论由分析论证和随机模拟得以支持。