【作者】 张红伟；

【导师】 袁赣南；

【作者基本信息】 哈尔滨工程大学 ， 导航、制导与控制， 2013， 博士

【摘要】 由于INS存在位置误差积累的问题，无法满足潜艇等水下潜器长期高精度的导航需求，因此为提高导航精度、获得高可靠性的位置信息，必须对INS进行定期修正，由此，重力场辅助导航定位系统应运而生。该系统目前还处于探索性阶段，若干关键性问题还需进一步地解决和完善。本文以重力场辅助导航定位技术为研究背景，以提高潜器的定位精度及可靠性为目的，对重力梯度基准图的构建、决策区域的确定和相关匹配解算算法的改进等关键技术展开研究工作。首先，在深入研究传统重力场辅助导航定位系统结构框架的基础上，分析总结该系统中每一个组成部分的重要性；针对重力梯度正演过程中地质密度模型无法获得的问题，着重研究利用现有的数据源来模拟地质密度并将其应用于变密度下的重力梯度正演算法中，给出一种利用高精度重力异常数据和数字高程数据在最小二乘法作用下推算地质密度的方法，将其与矩形棱柱法相结合以分析残余密度对重力梯度正演算法的影响。仿真结果表明，数字高程仍是影响重力梯度正演算法精度的主要因素，在正演过程中，陆地区域应该考虑应用较适合该区域的局部平均密度，而水下区域可使用剩余密度(1.643g/cm3)。针对目前误差椭圆方法不能很直观地表达出决策区域(搜索区域)的性能，且不便于程序实现的问题，以直接法和坐标变换法两种方式推导了由误差椭圆确定决策区域的方程表达式；详细讨论了误差椭圆和以“3σ”为原则的常规矩形方法之间的关系，得出当INS东向误差标准差大于北向误差标准差时，误差椭圆的相切矩形与常规矩形区域相重合，而当INS东向误差标准差小于北向误差标准差时，误差椭圆的相切矩形与常规矩形呈y=x对称的结论。对中低等精度的INS进行3组仿真实验，结果显示，误差椭圆相切于常规矩形，同时，每个真实位置点均包含在了误差椭圆区域内，从而证明以误差椭圆来确定决策区域方法的正确性。在详细推导和分析相关匹配算法中的等值线算法基础上，研究等值线算法在旋转和平移变换过程中的相关性问题，发现旋转和平移的计算具有严格的先后顺序，这样必然会将旋转矩阵的计算误差反作用于平移向量，出现混合误差的升高。针对该问题，为降低旋转误差传播对平移向量的影响，提出一种最近点加密的等值线改进方法，并对其进行仿真验证，结果表明，改进方法是有效的，在初始匹配误差较小的情况下，其定位精度优于传统的等值线算法。为提高重力异常辅助导航算法在大的初始匹配误差下的定位精度，详细分析基于概率神经网络的匹配算法，针对该算法对INS航迹过于依赖的缺陷，提出一种基于航迹自旋转微调的概率神经网络匹配算法。该算法由航迹初始化、航迹概率粗竞争、航迹概率精竞争和航迹概率最终竞争四部分组成，具有自动调节和获取最佳匹配航迹的特点。分别在4组具有典型特征的重力异常区域进行仿真实验，结果表明，改进的概率神经网络算法能够达到较高的匹配率，可以在很大程度上克服常规算法依赖于INS航迹的缺陷，并且继承了常规算法在大的初始匹配误差下依然能很好地工作的特点。为克服重力梯度辅助导航算法在大的初始匹配误差下定位精度不高的缺陷，将概率神经网络算法引入到重力梯度辅助导航中来，提出一种利用概率神经网络算法进行初始匹配位置调优的等值线改进方法：在决策区域内，利用概率神经网络算法对初始匹配位置进行调优来降低INS误差，从而形成待匹配航迹，在此基础上利用等值线算法得到最佳匹配位置。分别在不同初始匹配条件下进行仿真分析，得出概率神经网络算法在大的初始匹配误差下不易发散但定位精度不高的结论。分别在初始匹配误差为3.720′、4.617′和5.438′的条件下对改进方法进行仿真验证，结果表明，改进方法即使在大的初始匹配误差下仍然能够达到较高的定位精度。更多还原

【Abstract】 Due to the problem of INS position error accumulation, which is not possible to meetlong-term and high-precision navigation needs for the submarines and other underwatervehicles. So as to improve the navigation accuracy and obtain the high reliability locationinformation, the INS error must be periodically revised. As a result, the gravitational fieldaided navigation and positioning system came into being. The system is still in theexploratory stage, a number of key issues need to be further resolved and improved. Thegravity field aided navigation and positioning technology is as the research background inpaper, in order to improve the submersible navigation accuracy and reliability as the goal, thekey technology is researched, which is involved with the gravity gradient reference mapconstruction, the decision area determine and matching solution algorithm improved and soon.Firstly, the structure of the traditional gravitational field aided navigation and positioningsystem is studied in-depth. On the basis, the importance of each component part is analyzed.According to the problem that geological density model can not be obtained in gravitygradient forward process, the existing data sources are used to simulate the geological densityand applied to gravity gradient forward algorithm of the variable density. A method ofcalculating geological density based on high precision gravity anomaly data and digitalelevation data is given in the least squares estimation method. The method is combined withthe rectangular prism method to analyze the influence of the residual density on the gravitygradient forward algorithm. Simulation results show that the digital elevation is still the mainfactor that affects the accuracy of gravity gradient forward algorithm. In the gravity gradientforward process, the land area should consider applying the local average density, theunderwater area can apply the residual density (1.643g/cm3).According to the problem that the error ellipse method is not very intuitive to expressdecision area (search area) performance, and is not convenient to program implementation,the error ellipse equation expression is derived by the direct method and coordinatetransformation method. The relationship is discussed in detail between the error ellipse andthe conventional rectangular method to the "3σ " principle, the conclusions are obtained thaterror ellipse tangent rectangle coincides with conventional rectangular area when INS easterror standard deviation is greater than the north standard deviation, and error ellipse tangentrectangle and conventional rectangle are y=xsymmetrical when the INS east error standard deviation is less than the north standard deviation. In view of low and middle precision INSthe three simulation experiments are carried out, results show that the error ellipse is tangentto the conventional rectangular, at the same time, each real location points are contained in theerror ellipse area, thus the decision area method of determineing with the error ellipse iscorrect.The contour algorithm is derived in detail and analyzed in the correlation matchingalgorithm, on the basis, contour algorithm is studied on the correlation problem of therotational and translational transformation process. The problem is found that rotation andtranslation calculation has strict order, which inevitably rotation matrix calculation errorreacts on the translation vector, the mixing error enhance is caused. In order to solve theproblem, a closest point encryption contour improved algorithm is proposed. And thesimulation experiments are carried out, results show that the improved method is effective, thepositioning accuracy is better than the traditional contour algorithm when the initial matchingerror is small.In order to improve the navigation accuracy of the gravity anomaly aided navigationalgorithm in large initial matching error, probabilistic neural network matching algorithm isdiscussed in detail. According to the defect that the algorithm is too dependent on INS track, aprobabilistic neural network matching algorithm based on the track rotary trimming isproposed. The algorithm consists of four parts: track initiation, track probability roughcompetition, track probability sperm competition and track probability final competition, andthe algorithm is with automatic adjustment and geting the best matching track feature.Respectively the simulation experiments are carried out in four groups typical characteristicsgravity anomaly area, results show that the improved probabilistic neural network algorithmis able to achieve higher matching rate. The defect can be overcome that the conventionalalgorithm depends on the INS track and the characteristics is inherited that conventionalalgorithm is still able to work well in the large initial matching error.In order to overcome the defect that navigation accuracy of gravity gradient aidednavigation algorithm is not high in large initial matching error, the probabilistic neuralnetwork algorithm is introduced to gravity gradient aided navigation. At the same time, aimproved contour method is proposed that probabilistic neural network algorithm is used fortuning the initial position: in the decision area, probabilistic neural network algorithm is usedfor tuning the initial matching position to reduce INS error, and the matching track is formed.On this basis, the best matching position is got by contour algorithm. Respectively thesimulation analyses are done in different initial conditions, the conclusion is got that the probabilistic neural network algorithm is not easily divergence in the large initial error butnavigation accuracy is not high. Simulation experiments are carried out with the improvedmethod in the initial error3.720′、4.617′and5.438′, results show that the improved methodis still able to achieve a high navigation accuracy even in the large initial error.更多还原