航天器遥测时间序列数据挖掘研究

发布时间：2018-12-18 09:15

【摘要】：航天器作为航天事业的主要载体,是人类进行空间探索的基础。保证在轨航天器的正常工作,关系到整个航天工程的顺利执行。航天器遥测数据与地面模拟仿真实验得到的数据相比,更能反映航天器的真实工作状态,也更具可靠性,是航天器性能监测和实时状态分析的主要依据。有效利用在轨遥测数据并提取有效信息,不仅能为航天器管理决策提供支持,更能对航天器的设计改进起到参考作用。本文以某航天器2011年至2015年期间近200万行遥测数据为基础,针对遥测参数的特点,设计并实现遥测时间序列的特征表示、相似性度量以及中心序列计算这三个方面的算法。本文的主要工作和创新点如下:(1)针对航天器遥测数据参数众多,类型复杂的特点,设计了一种基于全局信息熵的自适应分段线性表示方法GIE-APLA。该方法弥补PLA方法在计算效率方面的不足,采用信息熵来度量当前数据段的波动,以达到线性时间内自适应划分的目的。在划分所得的子序列段中采用线性回归拟合原始序列,以保证特征表示的精度。实验结果表明,该算法在保证压缩率的前提下,对原始序列有较高的表示精度,为后续研究奠定了基础。(2)针对现有时间序列相似性度量方法的不足,提出了一种基于自适应线段的动态时间规整算法ASDTW。该算法针对DTW算法计算开销过大的问题,首先采用GIE-APLA算法将原始序列表示为序列段的形式,并根据其几何特征定义序列段之间的距离,在动态匹配阶段使用序列段作为基本匹配单元改善传统逐点匹配策略所导致计算开销过大的问题。实验结果表明,ASDTW算法保证度量精度的前提下,解决了DTW算法逐点匹配造成计算开销过大的问题。(3)针对现有中心序列算法计算开销过大和对合并顺序敏感的问题,提出了一种基于序列段的中心序列算法SSB。该算法首先通过层次聚类对序列集进行相似性的划分,以减少不同形态序列之间的影响;然后在各序列子集中以迭代的方式求解中心序列。考虑到迭代和动态匹配所造成的计算开销,在每次迭代过程中,使用序列段的匹配来减少计算规模,并通过定义序列段的质心来减少合并顺序对结果的影响。实验表明,SSB算法所得中心序列在表征能力上优于目前的NLAAF算法,与DBA算法相比性能持平;在计算效率上要优于上述两种算法。
[Abstract]:As the main carrier of spaceflight, spacecraft is the basis of human space exploration. Ensuring the normal operation of the orbiting spacecraft is related to the smooth implementation of the whole space project. Compared with the data obtained from the ground simulation experiment, the telemetry data can reflect the real working state of the spacecraft and be more reliable. It is the main basis for the performance monitoring and real-time state analysis of the spacecraft. The effective use of in-orbit telemetry data and the extraction of effective information can not only provide support for spacecraft management decision, but also play a reference role in spacecraft design improvement. Based on nearly 2 million lines of telemetry data from 2011 to 2015, this paper designs and implements three algorithms of telemetry time series, such as feature representation, similarity measurement and center sequence calculation, according to the characteristics of telemetry parameters. The main work and innovations of this paper are as follows: (1) aiming at the characteristics of many parameters and complex types of spacecraft telemetry data, an adaptive piecewise linear representation method based on global information entropy (GIE-APLA.) is designed. This method makes up for the deficiency of PLA method in computing efficiency, and uses information entropy to measure the fluctuation of current data segment, so as to achieve the purpose of adaptive partitioning in linear time. In order to ensure the accuracy of feature representation, linear regression is used to fit the original sequence in the subsequence segment. The experimental results show that the algorithm has a high precision for the original sequence under the premise of ensuring compression ratio, which lays the foundation for further research. (2) aiming at the shortcomings of the existing methods of measuring the similarity of time series, An adaptive line segment based dynamic time warping algorithm (ASDTW.) is proposed. In order to solve the problem of excessive computational overhead of DTW algorithm, the algorithm first uses GIE-APLA algorithm to represent the original sequence as the form of sequence segment, and defines the distance between sequence segments according to its geometric characteristics. In the dynamic matching phase, the use of sequence segments as the basic matching unit to improve the traditional point-by-point matching strategy leads to the problem of excessive computational overhead. The experimental results show that the ASDTW algorithm solves the problem that the point by point matching of the DTW algorithm leads to too much computing overhead. (3) the existing central sequence algorithm is too expensive and sensitive to the merging order. This paper presents a central sequence algorithm SSB. based on sequence segments. Firstly, the similarity of sequence sets is partitioned by hierarchical clustering to reduce the influence between different morphological sequences, and then the central sequence is solved iteratively in each sequence subset. Considering the computational overhead caused by iteration and dynamic matching, the matching of sequence segments is used to reduce the computational scale during each iteration, and the effect of merging sequence on the result is reduced by defining the centroid of sequence segments. The experimental results show that the center sequence obtained by the SSB algorithm is superior to the current NLAAF algorithm in representation ability and is equal to that of the DBA algorithm, and its computational efficiency is better than the above two algorithms.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：V557;TP311.13

【参考文献】