基于支撑向量机的卫星姿控系统异变特征提取

发布时间：2018-01-15 07:44

本文关键词：基于支撑向量机的卫星姿控系统异变特征提取　出处：《电子科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：目前,当今迅猛发展的科学技术中的尖端技术之一,航天技术正以最快的速度在越来越广的范围内,包括在政治、经济、军事、生活以及科学技术等众多的领域内发挥着非常卓越的作用。与此同时,航天的快速发展也使得各类航天器内部结构以及所需要的功能变的日益复杂,对于价格高昂的一些航天器,系统具有很高可靠性是整个航天器系统的最基本的需求,卫星系统是航天系统中很重要的一部分,一旦卫星的某个部件发生故障,轻的会使整颗卫星的提前设定的功能达不到预期目的或者直接丧失,非常严重的情况下,甚至可能导致发生一些严重灾难性的事件,还有这将会浪费国家巨大的财产。直到目前为止,对卫星的高可靠性的保证通常是通过确保软硬件的高的可靠性以及冗余来实现的。本文主要对卫星的下传状态数据进行了分析,提取出可以表征卫星当前状态的信息,判断卫星的运行状态。本文阐述了卫星姿控系统的主要组成部分,在动力学以及运动学两个方面分别对模型进行建模,并采用了四元数与欧拉角两种并行的方式对姿态信息进行详细的描述,为后续的算法的研究提供了基础。然后,采用了经验模态分解法提取信号的特征值并利用信号的能量熵进行阈值设定判断是否有异常发生,接下来,介绍了支撑向量机(SVM)的主要原理以及如何对两个影响比较明显的参数采取优化措施,提出了一种新的非线性的方式对参数进行优化,与现有的优化算法进行比较,它加快了参数的收敛速度,有效的降低了适应度值。运用SVM提出了两种方法提取卫星状态信息,第一是对卫星的状态进行建模,获取与实际观测值之间的残差信息,进行状态信息提取,并针对复杂数据样本情况,提出了一种优化的SVM回归方法,有效的改善了拟合效果并且支撑向量数较少,避免了过学习,第二是提出了一种对信号进行样本的符号化,并提取符号熵,对符号熵利用SVM进行分类识别运算,信号的符号熵值能有效的区分各种不同的状态,再利用SVM算法进行分类识别与现有方法比较提高了分类精度。
[Abstract]:At present, one of the cutting-edge technologies in the rapid development of science and technology, space technology is at the fastest speed in a wider and wider range, including political, economic, military. Life and science and technology play a very important role in many fields. At the same time, the rapid development of space also makes the internal structure of various spacecraft and the required functions become increasingly complex. For some expensive spacecraft, high reliability of the system is the most basic requirement of the whole spacecraft system, satellite system is a very important part of the space system, once a satellite component failure. The light can cause the entire satellite's predefined function to fail to achieve its intended purpose or to be directly lost, and in very serious cases, it may even lead to some serious catastrophic events. And it would be a waste of the country's huge assets. Until now. The guarantee of high reliability of satellite is usually realized by ensuring high reliability and redundancy of hardware and software. The main components of satellite attitude control system are described in this paper. The model is modeled in two aspects: dynamics and kinematics. And the quaternion and Euler angle are used to describe the attitude information in detail, which provides the basis for the subsequent research. The empirical mode decomposition method is used to extract the eigenvalue of the signal and the energy entropy of the signal is used to set the threshold to determine whether there is an anomaly or not. This paper introduces the main principle of support vector machine (SVM) and how to optimize the two parameters which have obvious influence, and puts forward a new nonlinear way to optimize the parameters. Compared with the existing optimization algorithm, it accelerates the convergence speed of the parameters and effectively reduces the fitness. Using SVM, two methods are proposed to extract satellite state information. The first one is to model the state of the satellite, obtain the residual information between the observed data and obtain the state information, and propose an optimized SVM regression method for the complex data samples. It can effectively improve the fitting effect and the number of support vectors is less, avoid overlearning. The second is to propose a signal sample symbolization, and extract symbol entropy. The symbol entropy is classified and recognized by SVM, and the signal entropy can effectively distinguish different states, and the classification accuracy is improved by using the SVM algorithm compared with the existing methods.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：V448.22

【参考文献】