基于BP神经网络的卫星激光测距望远镜指向误差的建模研究

发布时间：2018-07-29 17:24

【摘要】：随着光、机、电技术的迅速发展，卫星激光测距（SLR）技术已日渐成熟。在数据精度方面，无论国内外的固定站，还是流动站都已步入亚厘米级时代。近年来，随着高重复率激光技术的广泛应用，国内各SLR站在数据量和精度方面都进入了新阶段。但是，随着国内各SLR站在白天测距、盲目跟踪、无人值守观测和月球激光测距领域的探索和研究，，进而对卫星激光测距指向精度也提出更高的要求。在卫星激光测距观测中，提高望远镜的指向精度对提高跟踪目标的捕获能力具有重要意义，尤其在盲目跟踪或白天测距中，是保证观测数据获取的重要指标之一。常用的指向误差的建模方法有球谐函数模型、基本参数模型和转台模型。球谐函数模型表达式的各项没有物理意义，模型不稳定；基本参数模型的模型参数有实际的物理意义，模型比较稳定；转台模型是对基本参数模型的扩展，精度更高，但没有基本参数模型稳定。针对卫星激光测距望远镜指向误差问题，文本提出一种基于BP（Back Propagation）神经网络的建模方法。BP神经网络是一种按误差逆传播算法训练的多层前馈网络，是目前应用最广泛的神经网络模型之一。将BP神经网络算法应用到流动卫星激光测距仪的指向误差建模中，不仅拓展了神经网络算法的应用领域，同时对提高卫星激光测距系统的跟踪精度具有重要意义。通过获取武汉流动卫星激光测距站的三组实测恒星观测数据，将其分为训练集和检验集。利用训练集分别建立BP神经网络模型、转台模型、球谐函数模型和基本参数模型，再利用检验集对建立的模型进行测试，最后计算并对比四种模型输出的中误差。研究表明BP神经网络模型可以提高卫星激光测距望远镜的指向精度，且优于另外三种模型。另外，利用武汉流动卫星激光测距站的实测数据，并辅以长春SLR站和上海天文台SLR站的5组观测数据，进一步研究了指向误差的分布特性。首先采用遗传算法将武汉流动SLR站的观测数据分为训练集和检验集，利用转台模型建模。然后分析检验集样本残差的分布特性和指向精度。实验结果表明，转台模型输出的高度误差和方位误差的残差不一定符合正态分布，并且可通过遗传算法优化模型，提高望远镜的指向精度。最后，考虑到指向误差模型的实用性，并进一步完善指向误差模型，文章介绍了基于MATLAB开发环境，将各种模型建模方法的界面化过程。
[Abstract]:With the rapid development of light, machine and electric technology, the (SLR) technology of satellite laser ranging has matured day by day. In terms of data accuracy, both domestic and foreign fixed stations or mobile stations have stepped into the sub-centimeter era. In recent years, with the wide application of high repetition rate laser technology, all SLR stations in China have entered a new stage in terms of data volume and accuracy. However, with the exploration and research of SLR stations in the field of daytime ranging, blind tracking, unmanned observation and lunar laser ranging, the pointing accuracy of satellite laser ranging is also required. In satellite laser ranging observation, it is very important to improve the target acquisition ability by improving the pointing accuracy of the telescope. Especially in blind tracking or daytime ranging, it is one of the important indicators to ensure the acquisition of observation data. The commonly used modeling methods of pointing error include spherical harmonic function model, basic parameter model and turntable model. The expressions of spherical harmonic function model have no physical meaning and the model is unstable; the model parameters of the basic parameter model have practical physical significance and the model is relatively stable; the turntable model extends the basic parameter model and has higher precision. But no basic parameter model is stable. Aiming at the problem of pointing error of satellite laser ranging telescope, this paper presents a modeling method based on BP (Back Propagation) neural network. BP neural network is a multi-layer feedforward network trained by error back-propagation algorithm. It is one of the most widely used neural network models. The application of BP neural network algorithm to the modeling of pointing error of mobile satellite laser rangefinder not only expands the application field of neural network algorithm, but also plays an important role in improving the tracking accuracy of satellite laser ranging system. Three groups of observed star data from Wuhan mobile satellite laser ranging station were obtained and divided into training set and test set. The BP neural network model, the turntable model, the spherical harmonic function model and the basic parameter model are established by using the training set, and the test set is used to test the established model. Finally, the middle errors of the four models are calculated and compared. The results show that the BP neural network model can improve the pointing accuracy of the satellite laser ranging telescope and is superior to the other three models. In addition, using the measured data of Wuhan mobile satellite laser ranging station and five sets of observation data from Changchun SLR station and Shanghai Observatory SLR station, the distribution characteristics of pointing error are further studied. First, the observation data of Wuhan mobile SLR station are divided into training set and test set by genetic algorithm, and the turret model is used to model the data. Then, the distribution characteristics and pointing accuracy of the sample residuals in the test set are analyzed. The experimental results show that the height error and azimuth error of the turntable model are not always in accordance with the normal distribution, and the pointing accuracy of the telescope can be improved by optimizing the model by genetic algorithm. Finally, considering the practicability of the pointing error model and further improving the pointing error model, this paper introduces the interfacing process of various modeling methods based on MATLAB development environment.
【学位授予单位】：中国地震局地震研究所
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP183;P228

【参考文献】