基于数据挖掘技术的激变变星的特征提取

发布时间：2018-05-18 16:08

本文选题：数据挖掘 + 分类　；参考：《山东大学》2011年硕士论文

【摘要】：天体光谱中蕴含了丰富的物理信息,随着LAMOST望远镜投入使用,每个观测夜将获得上万条光谱。传统分析光谱的方法效率低、速度慢,无法满足对日益增长的数据的处理。数据挖掘作为信息发展到一定阶段的产物,从大量的、有噪声的的数据中提取出隐含在其中的有用信息,可以实现相关性预测、分类、聚类、孤立点发现、时间序列分析等许多功能,尤其是高维数据的处理非常有效。 LAMOST的海量天体光谱,不但对大样本天文学的研究有重要作用,还将产生许多副产品。数量如此巨大的光谱中隐藏着相对较多的稀有天体、时变天体和未知天体。激变变星是一种稀少天体,是研究吸积盘的“最佳天体物理实验室”,对其光学观测,特别是光谱观测对研究激变变星的物理特性和运动规律,对研究吸积盘的理论以及恒星演化等都具有重要意义。本文主要工作是通过研究激变变星的光谱,根据不同波段的特点,如巴尔默线系的发射与吸收、驼峰现象、双峰现象等,利用数据挖掘技术提取出已知激变变星的光谱特征,用于筛选激变变星的候选体。各类激变变星的光谱虽有一些共同特征,但是不同类型甚至同一类型的不同天体也有它的特殊性,加上某些爆发阶段的光谱与某些非激变变星光谱并无明显差别,因此,本文主要进行了以下工作： (1)研究了激变变星的主要特点,特别是其光谱特性,并且采用了PCA方法构造光谱的主分量,对光谱特征进行提取。采用主分量为轴,直接把样本点在主分量坐标轴上进行投影,可以得到二维平面上的样本特征点,大大降低了光谱数据的维数。 (2)分别研究了支持向量机、人工神经网络、K均值、K近邻等常用数据挖掘方法在分类和聚类上的应用。研究了一种新的方法：蚁群算法,分别就其在分类和聚类方而的模型进行研究 (3)在MATLAB不境下,根据数据挖掘的一般步骤,采用支持向量机、人工神经网络、K均值、K近邻以及随机森林,分别对同一数据集进行激变变星挖掘实验。对各种方法得到的结果进行时间、类CVs个数等综合分析和比较。比较不同方法筛选出激变变星的候选体,并分析其原因。
[Abstract]:There are abundant physical information in the spectrum of the celestial body. With the use of the LAMOST telescope, tens of thousands of spectra will be obtained each night. The traditional method of analyzing the spectrum is inefficient and slow, and can not meet the processing of growing data. Data mining is the product of information development to a certain stage, from a large number of noise. The useful information contained in it is extracted, which can realize correlation prediction, classification, clustering, isolated point discovery, time series analysis and many other functions, especially high dimensional data processing.
LAMOST's mass spectra of celestial bodies not only play an important role in the study of large sample astronomy, but also produce a number of by-products. In such a large spectrum, a relatively large number of rare objects are hidden, time-varying celestial bodies and unknown celestial bodies. The shock variable star is a rare celestial body and is the "best astrophysical laboratory" for the study of accretion discs. The optical observation, especially the spectral observation, is of great significance to the study of the physical characteristics and motion laws of the variable variable stars. It is of great significance to the study of the theory of the accretion disk and the evolution of the stars.
The main work of this paper is to study the spectrum of variable stars, according to the characteristics of different bands, such as the emission and absorption of the Ballmer line, the hump phenomenon, the Shuangfeng phenomenon and so on. The spectral characteristics of the known variable stars are extracted by data mining technology to select the candidate of the variable stars. Characteristics, but different types and even the same type of different celestial bodies have its particularity, and there is no obvious difference between the spectrum of some outbreak stages and some non - shock variation, so the following work is carried out in this paper.
(1) the main characteristics of the variable star are studied, especially its spectral characteristics, and the PCA method is used to construct the main component of the spectrum, and the spectral features are extracted. The sample point is directly projected on the principal component axis by using the principal component as the axis, and the sample feature points on the Er Weiping surface can be obtained, which greatly reduces the dimension of the spectral data. Number.
(2) the application of common data mining methods such as support vector machine, artificial neural network, K mean, K nearest neighbor and other common data mining methods in classification and clustering are studied. A new method, ant colony algorithm, is studied, which is studied on the model of classification and clustering respectively.
(3) under MATLAB, according to the general steps of data mining, using support vector machine, artificial neural network, K mean, K nearest neighbor and random forest, we carry on the experiment of variable star mining for the same data set, and analyze and compare the time and the number of CVs of the results obtained by various methods. The candidate of the variable star and its reasons.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TP311.13;P145.4

【引证文献】