半参数p-范极大似然回归

应用核权函数,在观测为误差单峰、对称的情况下,得到了一元p-范分布的半参数模型的计算公式.详细推导了p已知时一元p-范分布极大似然方程的解算公式,将半参数回归模型应用到极大似然平差的参数估计理论中,得到了一个比较好的算法.最后,构造了两个模拟平差问题,说明了此方法的优越性.     

Quantitative geomorphological analysis of drumlins in the Peterborough drumlin field,Ontario, Canada

Drumlins are enigmatic subglacial landforms that have been interpreted to form by a number of processes, including incremental accumulation of till, erosion of previously deposited sediment, catastrophic meltwater floods, and sediment deformation. However, relatively little is known about the controls on drumlin formation, such as spatially variable glacial processes or substrate characteristics, and how these controls may be identified from variations in drumlin morphology within a single drumlin field. This paper explores a computational method that allows identification of drumlins and extraction of their morphological characteristics from existing topographic digital data for a portion of the Peterborough drumlin field in Ontario, Canada. Spatial and non‐spatial analysis of the form and distribution of drumlins across the study area identifies drumlin characteristics such as size, elongation ratio, symmetry and long axis orientation and shows that drumlins are not randomly distributed across the region and their form characteristics have distinct regional trends. Kernel density analysis is used to identify the regional trends in drumlin characteristics. Factors that appear to influence the form and distribution of drumlins in the study area include sediment thickness, length of time beneath the ice, ice velocity and direction of ice movement. The distribution of particularly well developed asymmetric and elongate drumlins coincides with the location of a broad bedrock low and is interpreted to identify the former location of a fast‐flowing ice stream.     

基于半监督核模糊c-均值算法的北京一号小卫星多光谱图像分类

针对遥感图像数据大多不服从高斯分布以及遥感图像分类存在非线性、模糊性和标记数据少等问题,提出基于半监督核模糊c-均值算法的多光谱遥感图像分类方法.首先,把半监督学习理论和核理论同时引入模糊c-均值算法,形成半监督核模糊c-均值算法.然后,用该算法与k-均值算法、最大似然算法、多类支持向量、半监督核支持向量、模糊c-均值...     

基于球结构支持向量机的QuickBird影像分类分析

在支持向量机多类识别基础上探讨以球结构替代传统超平面支持向量机对QuickBird影像进行分类的可行性,对重叠区域的数据分类采用新规则,提高球结构支持向量机算法的泛化性能,并将分类结果与最小距离法、最大似然法分类结果进行比较,实验结果表明该算法有效可行,降低了二次规划的复杂度,缩短了样本训练时间.     

地震波数值模拟中的最优Shannon 奇异核褶积微分算子

如何有效协调地震波数值模拟过程中的计算精度和计算效率,一直是许多地球物理工作者密切关注且努力寻求解决的问题.这一问题求解的好坏,将直接影响到人们正确认识地下构造的地球物理响应.本文从离散Shannon奇异核理论出发,推导了基于Shannon奇异核的交错网格褶积微分算子,分析并求出了影响算子精度窗函数的最优化参数.通过将...     

Some modifications of Stokes' formula that account for truncation and potential coefficient errors

 Stokes' formula from 1849 is still the basis for the gravimetric determination of the geoid. The modification of the formula, originating with Molodensky, aims at reducing the truncation error outside a spherical cap of integration. This goal is still prevalent among various modifications. In contrast to these approaches, some least-squares types of modification that aim at reducing the truncation error, as well as the error stemming from the potential coefficients, are demonstrated. The least-squares estimators are provided in the two cases that (1) Stokes' kernel is a priori modified (e.g. according to Molodensky's approach) and (2) Stokes' kernel is optimally modified to minimize the global mean square error. Meissl-type modifications are also studied. In addition, the use of a higher than second-degree reference field versus the original (Pizzetti-type) reference field is discussed, and it is concluded that the former choice of reference field implies increased computer labour to achieve the same result as with the original reference field. Received: 14 December 1998 / Accepted: 4 October 1999     

非线性支持向量机在油气预测中的应用

油气预测的传统方法通常是基于经验风险最小化准则,但在有限样本情况下,预测效果并不理想.研究引入基于结构化风险最小化准则的非线性支持向量机方法,通过对推广误差界的最小化达到最大的泛化能力和全局最优,时于小样本数据,该方法具有可靠的预测能力.在对四川观音场构造面新统上部碳酸盐岩储层数据处理中.通过实例试算,结果表明该方法有效可靠,预测精度高,与已知结果吻合较好.     

基于核最小二乘模型的矿产靶区预测

地质统计单元的含矿性与地质找矿证据之间存在复杂的非线性关系,建立这种复杂关系的多元非线性统计模型并预测矿产靶区,对矿产勘查具有重要指导意义。以核函数为理论工具,在核最小二乘原理基础上提出了矿产靶区预测的核最小二乘模型;在GDAL数字图像输入输出函数库和CLAPACK线性代数软件包基础上,用VC++语言开发了面向栅格数据的矿产靶区预测核最小二乘模型算法程序,并把模型应用于新疆阿勒泰地区的矿产靶区预测研究。在MapInfo中生成包含100×151个网格统计单元的栅格图层,把栅格化后的15种找矿证据图层转化成100×151×15的数字图像数据立方体,用自行开发的程序计算每个网格统计单元的核最小二乘判别得分。结果表明,网格统计单元判别得分的高值区与已知矿床（点）的空间分布基本一致。     

一种基于核学习的储集层渗透率预测新方法

基于核学习的支持向量机,是一种采用结构风险最小化原则代替传统经验风险最小化原则的新型统计学习方法,具有完备的理论基础。这里提出了核学习技术在储集层非均质特性描述中渗透率参数预测的新用途。在复杂地层中,基于支持向量机的智能和自适应模式识别能力而建立了常规测井多参数信息输入的渗透率预测模型,然后对实际油田储集层渗透率进行了预测。与常规线性回归模型预测结果相对比,所提出的方法更易于使用,很少受不确定因素的影响,并具有较强的信息整合能力以及更高的预测准确性和可信度。     

Treatment Effects of Tunnel Entrance Collapse Section Based on Field Monitoring Results北大核心CSCD

为评价隧道洞口塌方段的安全性,基于累计变形预测和变形速率趋势分析模型,结合工程实例,研究隧道洞口塌方段的处理效果。实例分析表明：在累计变形预测方面,GA-VMD模型在隧道变形数据的信息分解过程中具有明显的优越性,可利用其将隧道变形数据分解为趋势项和误差项,且通过组合预测能有效保证预测精度,得出累计变形的后期仍具一定增加趋势,但增加速率相对较小;在变形速率趋势分析方面,变形速率具下降趋势,且随时间持续,下降趋势更为显著。综合对比两者结果,隧道变形无明显增加趋势,并趋于稳定方向发展,说明塌方处理措施是合理有效的,为其防治效果评价提供了理论依据。