LAMOST M型星光谱自动处理方法的研究

As the main components of the Milky Way, M dwarfs are long-lived and stable among the main sequence stars. Therefore, they are good tracers of the structure and evolution of the Milky Way. LAMOST has accumulated a large spectral sample of M dwarfs, from which obtaining accurate and reliable spectral parameters is the primary task for carrying out the relevant scientific research. In this project, spectra that were classified to M type by LAMOST 1D pipeline are automatically processed, the output result is a reliable and complete parameters catalog. The main contents are: (1) the outlier mining method is applied to remove a small amount of other spectral-type spectra and flux-anomaly spectra (2) identifies and labels M giants, M subdwarfs and M dwarfs.(3) Searching the optimal subset of spectral features, which is used to develop anti-noises method to calculate spectral parameters. Reliability estimation for the important parameters are provided. (4)The algorithms are parallelized into a software system that can automatically process spectrum of M-type star efficiently. Results of the project can provide convenience for scientific research in selection of pure samples and obtaining reliable parameters; mining a large number of spectrum of M-type stars and analyzing their spectral features may find hidden patterns and acquire a better understanding of M-type stars.

M矮星是银河系的主要构成成分，具有漫长而稳定的主序寿命，适宜用来开展银河系的结构和演化的研究。LAMOST已经积累了M矮星光谱大样本，获得准确可靠的光谱参数是开展相关科学研究的首要任务。本课题采用机器学习和数据挖掘算法自动处理LAMOST 1D Pipeline分类为M型的光谱，输出可靠完备的参数星表。研究内容包括：（1）利用离群点挖掘方法去除样本中少量的其他光谱型光谱和流量异常光谱（2）对样本中的M巨星、M亚矮星以及M矮星进行识别和标识（3）研究基于最优特征子集的抗噪声方法，用来计算光谱参数；对重要的参数给出可靠性估计。（4）对算法并行化编程，形成高效的M型星光谱自动处理软件系统。课题成果能为相关科学研究在选择纯净的样本、获得可靠的参数方面提供便利；研究中对大量M型星光谱的挖掘和特征分析有利于发现隐含的模式，加深对M型星的认识。

本课题在现有的LAMOST M矮星参数测量方法的基础上，研究运用机器学习、数据挖掘的算法，自动处理LAMOST pipeline分类为M的光谱，识别、排除非M型星光谱、M巨星光谱、M亚矮星光谱，获得较纯净的M矮星样本，并给出准确的光谱型、视向速度、各种分子带和原子线指数、磁活动强度等参数。研究结果包括：（1）研究对比了模板匹配、SVM、Random Forest、XGBoost、LightGBM、CNN及Wide&Deep等方法应用在光谱型分类和光度级分类上的效果，获得了分类准确率高的模型。（2）研究并掌握了M巨星和M亚矮星光谱的特征识别方法、特征重要性量化的方法。（3）得到具有较强抗噪声能力的、准确率高的光谱型和视向速度测量方法。（4）编制了一套自动处理LAMOST M型星光谱、获得参数的软件系统。（5）发表SCI和EI论文共10篇。该课题的研究成果以及编制的软件系统对于下一步测量M矮星的大气参数、M矮星的空间运动参数，从而构建M矮星的完备参数星表奠定了基础。