半参数回归模型虚拟1.docVIP

半参数模型解算的虚拟观测法 本项目由湖南省自然科学基金项目(02JJY2066)和湖南省科技计划项目联合资助(2004022200611) 朱建军　冯光财　戴吾蛟 中南大学信息物理工程学院 摘要 半参数模型中的非参数部分可以很好地描述测量数据处理规律不是十分明确的系统误差或模型误差，因而近年得到了测绘工作者的广泛重视。但目前半参数模型的各种解算方法主要还是沿用数学中提出的方法，例补偿最小二乘法，样条函数法，核光滑估计等，这些方法的特点是：所用的参数和语言都是纯数学的、相对抽象的，与具体应用中的实际意义关系不大，如何根据具体的问题确定方法中的有关量，没有成熟可靠的方法。本文首先介绍半参数回归中常用的补偿最小二乘法。然后基于先验信息，从纯测量学的观点讨论半参数模型的解算。即将对问题的先验信息转换成对问题的虚拟观测，用虚拟观测与原观测联合按常规的最小二乘方法求解。理论和实际都证明，该方法与最小二乘补偿法完全等价。从而在理论上得到一个重要的结论：半参数回归的补偿最小二乘法中的正则矩阵可由虚拟观测的观测方程系数确定，即，，平滑因子可由观测方差与虚拟观测方差的方差比（权比）确定，而该方差比可以在计算中用方差分量估计的方法确定。由此将半参数回归的解算与传统的测量数据处理方法有机地结合起来了。实例的计算结果表明，本文提出的虚拟观测方法计算的结果一般要优于常规的补偿最小二乘结果，基本上可达到常规补偿最小二乘法在理论上的最优解。 关键词：半参数模型，补偿最小二乘法，先验信息　虚拟观测 中国图书分类号：P207 A quasi observation approach for semi-parameter regression Zhu Jianjun Feng Guangcai Dai wujao School of info-physcs and Geomatics, Central Abstract The non-parameter in semi-parametric model can be used to describe the systemic error or model error in geodesy, so semi-parameter regression get a great attention in geodesy. However, all the methods to find solutions of the model are based on mathematics, some concepts in the methods are very abstract, especially some quantities are no relationship to practical situation. So it is very difficult to determine these quantities on practical situations. In this paper the method of penalized least squares(PLS) is introduced at first. And then, the solution of the model is studied on the view of geodesy. It is suggested that the prior information on the semi-parameters is transformed into quasi observations, and the quasi observations are adjusted then together with the real observations. The paper proves that the quasi observations method is equal to PLS. The regular matrix in PLS can be determined by coefficient matrix of the quasi observation equation, that is, , the smoothing parameter is equal to the ratio of the observation variance to the quasi observation variance. The example shows that the quasi observation method usually will be better than PLS. And it can get nea