【作者】 胡伍生；

【导师】 吴中如； 华锡生；

【作者基本信息】 河海大学 ， 水工结构工程， 2001， 博士

【摘要】 本文系统地研究了在地壳形变监测中GPS的精密高程测量技术及信息处理的有关理论和方法。主要研究内容概述如下： （1）简要介绍了GPS定位的基本原理、GPS精密定位的方法及其误差来源分析。差分GPS技术能有效地消除很多物理因素对GPS测量结果的误差，但却不能十分有效地消除大气对流层折射对GPS高程测量的影响。因此，如何处理对流层折射影响就成为提高GPS高程测量精度的关键。作者详细研究了模拟对流层折射影响的随机过程方法：一阶高斯马尔可夫模型和分段线性模型。这两种方法能十分有效地模拟对流层折射的影响，从而使GPS高程测量的精度大大提高。分段线性方法的结果与附加参数的选取个数有关，经过某工程实例的试验研究发现，取2h～4h附加一个参数来模拟对流层折射的影响，效果比较理想。 （2）详细介绍了神经网络及BP算法的基本理论，并针对BP算法存在的问题，作者提出了相应的改进措施。以一个简单的三层BP网络模型“Model［2×2×2］”为例，推出了BP算法正向、反向过程中的所有计算公式。作者提出了用神经网络方法来转换GPS高程的五层BP网络结构，并对该BP网络的具体结构模型进行了详细的试验研究，得出了一个最佳的BP网络结构模型。转换GPS高程通常使用二次曲面拟合法和神经网络方法，但这两种方法各有优缺点。在二次曲面拟合法和神经网络方法的基础上，作者创新了一种转换GPS高程的新方法——“混合转换法”（简记为CF＆NNM方法），该方法思路独特新颖，在工程实践应用中效果良好。作者还对该方法的实质进行了解析，与二次曲面拟合法相比，利用CF＆NNM法转换GPS高程，其转换结果精度大约可提高20％，具有推广价值。另外，作者又创新了一种对BP网络计算的改进算法——“误差分级迭代法”，采用该法能加快BP网络训练的收敛速度，能有效地克服初始权值对模拟结果的影响，还能有效地减弱样本次序对BP网络训练结果的影响，从而改善了BP网络的性能。 （3）在地壳形变动态模型的研究中，作者建立了动态间接平差模型，推导了其主要平差计算公式，并讨论了它与静态分期平差之间的关系。作者还推导出了水准网和GPS网进行联合动态平差的观测值误差方程式的通用公式，经实例检验，该公式方便实用。与此同时，还对Helmert验后方差迭代定权的严密公式进行了简化，简化后的公式计算简单，迭代定权收敛速度快，并对动态平差模型下形变显著性的统计检验作了详细研究。 （4）介绍了作者开发的基于Access2000和Visual Basic 6.0的地壳垂直形变监测信息管理系统。利用该系统可自动生成各类成果报表、自动绘制沉降曲线图等。更多还原

【Abstract】 In this dissertation, the theory and the method of GPS precise positioning and data processing in monitoring the earth’s crust deformation are systematically discussed. The main research contents of this dissertation are summarized as follows: (1) The basic theory of GPS positioning~ the method of GPS precise positioning and the source of its error are briefly introduced in the paper. Difference GPS technique can effectively eliminate the error of many physics?factors to GPS result, but can not effectively eliminate the error of the tropospheric refraction to GPS height. Therefore, the key to improve the measuring accuracy of GPS height is how to deal with the error of the tropospheric refraction. The methods of stochastic process (such as one-order model of Guess-Marlkov, the piece wise linear method) used for simulating the error of the tropospheric refraction are studied in detail. Both methods above can effectively eliminate the error of the tropospheric refraction to GPS result, and can obviously improve the accuracy of GPS height. The result by the piece wise linear method bears relation to the numbers of additional parameter in adjustment. After being tested and studied with one project, it is proved that the result of adjustment is good, if we use one additional parameter per 2 hours or per 4 hours to simulate the error of the tropospheric refraction. (2) The basic theory of neural network and an algorithm of BP are described in this paper. Many improvement steps to the problems existed in the algorithm of BP are put forward by the author. Putting a simple model (Model [2× 2×2]) of three-layer BP network as an example, all the formula for the algorithm of BP are deducted. A new structure of five-layer BP network used for converse GPS height is put forward in this article, and the optimal structure of the BP network is acquired after hundreds of tests. Normally, the conicoid fitting method (CBt) and the neural network method (NNM) are used for converting a GPS height into a normal height. But, each of them has its own advantages and disadvantages. After studying these two methods, a new method named Mixed-ethod (abbr. CF&\NM) is conceived that combines the advantages of both the conicoid fitting (CFM) and neural network method (NNM). The thought of the Mixed-method is an originality idea, and the effect of the new method is very good in the application of some projects, then the essence of the Mixed-method is analyzed. Comparing with the conicoid fitting method (CFM) to converse GPS height, the accuracy of the transform result with the CF&NNM method can improve 20 percent. The new method is worthy of extending. Otherwise, a betterment algorithm of BP network, named “error grade iterative method”,is innovated by the author. By using the betterment method, the study convergence speed of network can be accelerated, the error of the simulating result influenced by the first weight and by the sequence of samples can be overcome, and the performance of BP network can be improved. (3) In the research of dynamic model of the erustal deformation, a model of kinematic adjustment of indirect observations has been established in this paper, its main adjustment formula has been deduced, and the relationship bet-ween static separate adjustment and the kinematic adjustment has been discussed. The current formula of the dynamic adjustment’s error equation of observations united level net and GP更多还原