近些年海量恒星光谱数据的获取使得恒星三个基本参数（表面有效温度Teff, 表面重力log g, 金属丰度［Fe/H］）的自动测量方法研究成为一个重要的研究课题, 相应对恒星大气物理参数测量的研究将对科学家研究宇宙演化等重大科学问题具有重要的意义。 但是目前国内外针对此问题所做的研究并不是很广泛深入, 且已有的一些方法还不能够完全准确地估计出恒星的大气物理参数。 因此本文研究了一种基于质量估计的恒星大气物理参数自动测量方法, 该方法计算量比较小, 其主要思想是首先建立一些质量分布, 将原始光谱数据经过质量估计算法映射到新的质量空间, 然后在质量空间利用支持向量机回归对恒星的三个基本物理参数进行估计。 在实验中, 从SDSS-DR8光谱数据库中选择部分实测光谱数据来进行训练和测试, 并将该方法预测出的参数结果与SSPP给出的参数值进行了对比, 取得了比较理想的结果。 实验结果表明, 该研究方法的准确率更高, 预测结果更稳定, 训练所用的时间短, 在恒星大气物理参数自动测量上是行得通的, 可以有效地测量恒星的大气物理参数。

We have collected massive stellar spectral data in recent years, which leads to the research on the automatic measurement of stellar atmospheric physical parameters (effective temperature Teff, surface gravity log g and metallic abundance ［Fe/H］) become an important issue. To study the automatic measurement of these three parameters has important significance for some scientific problems, such as the evolution of the universe and so on. But the research of this problem is not very widely, some of the current methods are not able to estimate the values of the stellar atmospheric physical parameters completely and accurately. So in this paper, an automatic method to predict stellar atmospheric parameters based on mass estimation was presented, which can achieve the prediction of stellar effective temperature Teff, surface gravity log g and metallic abundance ［Fe/H］. This method has small amount of computation and fast training speed. The main idea of this method is that firstly it need us to build some mass distributions, secondly the original spectral data was mapped into the mass space and then to predict the stellar parameter with the support vector regression (SVR) in the mass space. we choose the stellar spectral data from the United States SDSS-DR8 for the training and testing. We also compared the predicted results of this method with the SSPP and achieve higher accuracy. The predicted results are more stable and the experimental results show that the method is feasible and can predict the stellar atmospheric physical parameters effectively.