An analysis of constrained robust regression estimators

Multicollinearity and nonnormal errors are problems often encountered in the application of linear regression. Estimators are proposed for dealing with the simultaneous occurrence of both multicollinearity and nonnormality. These estimators are developed by combining biased estimation techniques with certain robust criteria. An iteratively reweighted least-squares procedure is used to compute the estimates. The performance of the combined estimators is studied empirically through Monte Carlo experiments structured according to factorial designs. With respect to a mean-squared-error criterion, the combined estimators are superior to ordinary least-squares, pure biased estimators, and pure robust estimators when multicollinearity and nonnormality are present. The loss in efficiency for the combined estimators relative to least squares is small when these problems do not occur. Some guidelines for the use of these combined estimators are given.     

基于EV模型的惯导平台误差分离方法

针对惯导平台误差分离过程中输入输出观测数据均含有噪声的问题,应用基于EV模型的总体最小二乘方法进行误差分离.给出了惯导平台的误差模型及误差观测方程,介绍了EV模型及总体最小二乘方法,分析了利用车载试验进行误差分离时平台的安装方式以及所需的外测信号等问题.应用最小二乘法和总体最小二乘法进行仿真对比研究.仿真结果表明,基于EV模型的总体最小二乘法对惯导平台的误差系数分离精度较最小二乘法要高.     

滑动容错辨识算法在航迹数据处理中的应用

由于试验环境和被测目标运动特性的影响,航迹测量数据中会出现斑点型异值。对最小二乘多项式进行改进,引进自适应调节函数和滑动窗口概念,提出了滑动多项式容错辨识算法,详细描述了采用该方法进行异值辨识和修复的具体过程。仿真分析和实际飞行试验的数据处理表明:该方法对处理含有斑点型异值的测量数据具有良好的效果。     

基因表达缺失值的加权回归估计算法

在基因芯片实验中,数据缺失客观存在,并在一定程度上影响芯片数据后续分析结果的准确性。在不增加实验次数的情况下,缺失值估计是降低缺失数据对后续分析影响的有效方法。利用相似性信息的核加权函数来实现缺失值回归估计的局部化,提出了基于加权回归估计的基因表达缺失值估计算法。在两个不同类型的基因芯片数据上,将新方法与几种已知的方法进行了比较分析。实验结果表明,新的估计算法具有比传统缺失值估计算法更好的稳定性和估计准确度。     

Challenges of measuring progress in Afghanistan using violence trends: the effects of aggregation,military operations,seasonality, weather,and other causal factors

Measuring nationwide progress of counterinsurgency operations in Afghanistan using violence trends is difficult due to several factors: aggregation of data to the national level may obfuscate disparate local trends; the observed seasonality in violence makes comparisons difficult and may obscure progress; and short-term spikes or troughs – attributable to weather, military operations and tempo, or holiday periods – heavily influence simple averaging schemes. Despite these challenges, proper understanding of violence statistics is critical to estimating the effectiveness of military forces added during a surge or redeployed as part of transition. This article explores methods for analyzing observed violence trends to identify causal factors, to provide a comparable baseline, and to inform assessments at appropriate levels of aggregation. One methodology for seasonal adjustment of violence data is discussed and shown to provide a logical baseline for examining trends. An ordinary least squares regression model is developed and implemented using time-series violence data.     

Coordinated replenishments from multiple suppliers with price discounts

In this study we present an integer programming model for determining an optimal inbound consolidation strategy for a purchasing manager who receives items from several suppliers. The model considers multiple suppliers with limited capacity, transportation economies, and quantity discounts. We propose an integrated branch and bound procedure for solving the model. This procedure, applied to a Lagrangean dual at every node of the search tree, combines the subgradient method with a primal heuristic that interact to change the Lagrangean multipliers and tighten the upper and lower bounds. An enhancement to the branch and bound procedure is developed using surrogate constraints, which is found to be beneficial for solving large problems. We report computational results for a variety of problems, with as many as 70,200 variables and 3665 constraints. Computational testing indicates that our procedure is significantly faster than the general purpose integer programming code OSL. A regression analysis is performed to determine the most significant parameters of our model. © 1998 John Wiley & Sons, Inc. Naval Research Logistics 45: 579–598, 1998     

多视图三角化中特征点噪声尺度的自适应估算

鲁棒性多视图三角化方法通常借助重投影误差经验阈值来剔除图像对应中的错误匹配,该经验阈值的选取直接影响三维重构场景点的数量和精度。在分析图像特征点定位噪声及对极传递几何原理的基础上,建立对极传递过程不确定性的传递模型,提出一种基于核密度估计的最优噪声尺度估算方法,并将该噪声尺度作为多视图三角化中错误匹配筛选的依据。实验结果表明,该方法可以获得准确的噪声尺度估计,从而有效提升多视图三角化方法的三维重构质量。     

面向监督学习的稀疏平滑岭回归方法

岭回归是监督学习中的一个重要方法,被广泛用于多目标分类和识别。岭回归中一个重要的步骤是定义一个特殊的多变量标签矩阵,以实现对多类别样本的编码。通过将岭回归看作是一种基于图的监督学习方法,拓展了标签矩阵的构造方法。在岭回归的基础之上,进一步考虑投影中维度的平滑性和投影矩阵的稀疏性,提出稀疏平滑岭回归方法。对比一系列经典的监督线性分类算法,发现稀疏平滑岭回归在多个数据集上有着更好的表现。另外,实验表明新的标签矩阵构造方法不会降低原始岭回归方法的表现,同时还可以进一步提升稀疏平滑岭回归方法的性能。     

Prediction with zero-one loss structure

This paper poses a prediction problem in which a linear model is assumed. With a “zero-one” loss structure as the loss from incorrect prediction, it is suggested that least squares may not be appropriate for estimating the parameters of the model. An alternate criterion is proposed and integer programming is used in order to find the estimates, given the proposed criterion.     

矢量多边形并行栅格化数据划分方法

针对多边形并行栅格化中的负载不均衡问题提出一种新的数据划分方法,主要包括:迭代计算划分线的位置,在每次迭代中保证分块间的计算量大致均衡,完成数据划分、实现负载均衡;提出基于二叉树的划分结果融合策略,以解决跨边界多边形的融合问题。在多核CPU环境下实现并行算法,选用多个典型土地利用现状数据集进行测试。结果表明:针对不同类型多边形数据集,所提方法较传统方法可获得更高的并行加速比和更好的负载均衡;针对大数据量数据集,以多边形节点数为度量标准可更精确地估算分块计算量,从而更好地实现负载均衡。     

基于ANFIS的目标毁伤等级预测模型

分析了神经网络和模糊推理系统的优缺点,研究了自适应神经模糊推理系统（ANFIS）结构模型及后向传播和递归最小二乘算法相结合的混合算法．在分析了目标毁伤等级主要影响因素的基础上,构建了目标毁伤等级预测ANFIS模型,利用毁伤试验样本数据训练该模型,得到了与实际一致的目标毁伤等级,并将预测结果与基于BP神经网络的预测结果进行了仿真对比分析．仿真结果表明,该目标毁伤等级预测模型能够准确地预测出目标的毁伤等级,并且其预测精度较BP神经网络方法高,为目标毁伤等级预测提供了一种有效的方法．     

武器系统模糊寿命周期费用模型

武器系统的寿命周期费用建模较多采用参数法,而参数法中最常用的是最小二乘回归.考虑费用统计数据的模糊性,提出用模糊最小二乘回归来建立武器系统模糊寿命周期费用模型,并结合实例对武器系统寿命周期费用进行了分析.结果表明,这种方法能达到令人满意的拟合精度,具有实用价值.     

Estimating reliability growth (or deterioration) using time series analysis

In this paper we propose a method for estimating reliability growth (or deterioration) using time series analysis. Our method does not call for the specification of a particular model, and estimates the growth in the presence of periodicity. We illustrate our procedure by considering some binomial failure data generated during the testing of a large system of the U.S. Navy.     

Robust adaptive UKF based on SVR for inertial based integrated navigation

Aiming at the problem that the traditional Unscented Kalman Filtering (UKF) algorithm can’t solve the problem that the measurement covariance matrix is unknown and the measured value contains outliers, this paper proposes a robust adaptive UKF algorithm based on Support Vector Regression (SVR). The algorithm combines the advantages of support vector regression with small samples, nonlinear learning ability and online estimation capability of adaptive algorithm based on innovation. Firstly, the SVR model is trained by using the innovation in the sliding window, and the new innovation is monitored. If the deviation between the estimated innovation and the measured innovation exceeds a given threshold, then measured innovation will be replaced by the predicted innovation, and then the processed innovation is used to calculate the measurement noise covariance matrix using the adaptive estimation algorithm. Simulation experiments and measured data experiments show that SVRUKF is significantly better than the traditional UKF, robust UKF and adaptive UKF algorithms for the case where the covariance matrix is unknown and the measured values have outliers.     

基于压缩感知原理的融合判别信息的协作表示方法

提出了一种用于视觉分类任务的低计算复杂度且有效的图像表示方法。把协作表示和判别信息结合在统一框架内,是基于协作表示分类方法的一种扩展形式。测试样本的协作表示系数是稀疏的,这种基于冗余和过完备的表示对于遮挡和伪装而言是鲁棒的;此外,通过最小化类内散布矩阵和最大化类间散布矩阵的判别信息的挖掘,对于视觉分类问题也是很有帮助的。在一些基准数据库上的实验表明,提出的方法相对于现有的方法而言能够获得更有竞争力的表现。     

基于偏最小二乘回归的军用飞机采购价格预测

考虑到军用飞机采购价格样本数据少、难于预测的特点和偏最小二乘回归方法在处理小样本多元数据方面的优势,提出一种基于偏最小二乘回归的军用飞机价格预测方法.偏最小二乘回归首先提取第一、第二主成分对采购价格样本的特异点进行剔除;然后进行变量投影重要度分析来筛选变量;最后,偏最小二乘回归对筛选的变量进行回归建立军用飞机价格预测模型,并对军用飞机价格进行预测.结果表明,在军用飞机价格预测方面,与未筛选变量的回归模型和逐步多元回归相比,经过变量筛选的偏最小二乘回归模型预测的精度更高,更能体现采购价格与飞机性能参数之间的关系.     

Analysis of data from life-test experiments under an exponential model

This paper discusses situations in which the distribution of a lifetime response variable T is taken to depend upon a vector x of regressor variables. We specifically consider the case in which T, given x , has an exponential distribution, and in which x represents levels of fixed factors in an experimental design. Methods of analyzing data under this type of model are discussed, with maximum likelihood and least squares methods being presented and compared.     

基于多源冲突数据聚类的态势估计方法

传统调和式态势估计方法在面对多源冲突数据时融合效果不佳。为此,提出一种基于冲突数据聚类的非调和式态势估计方法。首先利用迭代自组织数据聚类方法(ISODATA)对多源冲突数据进行聚类,然后利用频度和可信度对数据簇的重要性进行评估,最后得到态势估计结果。仿真结果表明,与传统态势估计方法相比,所提方法在融合多源冲突数据时能够得到可信度较高的态势估计结果。     

Expressions for item fill rates in periodic inventory systems

We examine the problem of estimating the item fill rate in a periodic inventory system. We show that the traditional expressions for line item fill rate, found in many operations management textbooks, perform well for high fill rates (above 90%), but they consistently underestimate the true fill rate. The problem of underestimation becomes significant as the fill rate falls below 90% and is greatly amplified in cases with very low fill rates (below 50%). We review other more accurate expressions for fill rate, discussing their relative merits. We then develop an exact fill rate expression that is robust for both high and low fill rates. We compare the new expression to others found in the literature via an extensive set of simulation experiments using data that reflect actual inventory systems found at Hewlett-Packard. We also examine the robustness of the expressions to violations in the underlying assumptions. Finally, we develop an alternative fill rate expression that is robust for cases of high demand variability where product returns are allowed. © 1995 John Wiley & Sons, Inc.     

A nonparametric approach to accelerated life testing under multiple stresses

Accelerated life testing (ALT) is concerned with subjecting items to a series of stresses at several levels higher than those experienced under normal conditions so as to obtain the lifetime distribution of items under normal levels. A parametric approach to this problem requires two assumptions. First, the lifetime of an item is assumed to have the same distribution under all stress levels, that is, a change of stress level does not change the shape of the life distribution but changes only its scale. Second, a functional relationship is assumed between the parameters of the life distribution and the accelerating stresses. A nonparametric approach, on the other hand, assumes a functional relationship between the life distribution functions at the accelerated and nonaccelerated stress levels without making any assumptions on the forms of the distribution functions. In this paper, we treat the problem nonparametrically. In particular, we extend the methods of Shaked, Zimmer, and Ball [7] and Strelec and Viertl [8] and develop a nonparametric estimation procedure for a version of the generalized Arrhenius model with two stress variables assuming a linear acceleration function. We obtain consistent estimates as well as confidence intervals of the parameters of the life distribution under normal stress level and compare our nonparametric method with parametric methods assuming exponential, Weibull and lognormal life distributions using both real life and simulated data. © 1998 John Wiley & Sons, Inc. Naval Research Logistics 45: 629–644, 1998