基于线指数的核偏最小二乘回归在恒星大气物理参数测量中的应用

发布时间：2018-03-21 20:50

本文选题：Lick线指数　切入点：核偏最小二乘回归(KPLSR)　出处：《光谱学与光谱分析》2014年03期 　论文类型：期刊论文

【摘要】：恒星大气物理参数(有效温度、表面重力、化学丰度)的自动测量是天体光谱数据自动处理中的一项重要内容。由于光谱数据的高维性的特点,处理运算量非常大,对于光谱的实时分析及处理会造成延误。文章提出了一种基于Lick线指数,利用核偏最小二乘回归(KPLSR)对恒星大气物理参数进行测量的方法。可以有效地减少运算量并可达到理想的准确率。首先计算Kurucz合成光谱的Lick线指数,利用核偏最小二乘回归方法建立Lick线指数与大气物理参数之间的核回归模型,并利用DR8实测光谱数据对得到的模型进行测试,将测试的结果与SEGUE SSPP提供的大气物理参数进行了对比,取得了比较好的效果。此外,为了检验噪声对参数测量的影响,本文还对Kurucz光谱分别加了信噪比为10,20,30,40,50,70,90,120的高斯白噪声,对得到的不同信噪比的Kurucz数据进行了测试,实验结果表明,核回归模型对噪声比较敏感,光谱数据的信噪比越高,其大气物理参数的预测精度越高。提出的基于线指数建立核偏最小二乘回归模型的方法运算量小,训练速度快,适合用于恒星大气物理参数的测量。
[Abstract]:The automatic measurement of stellar atmospheric physical parameters (effective temperature, surface gravity, chemical abundance) is an important part of automatic processing of celestial spectral data. The real-time analysis and processing of spectrum will cause delay. In this paper, a new method based on Lick line exponent is proposed. The kernel partial least square regression (KPLSR) is used to measure the physical parameters of stellar atmospheres, which can effectively reduce the computation and achieve the ideal accuracy. Firstly, the Lick line exponent of the Kurucz synthesis spectrum is calculated. The kernel regression model between the Lick line index and the atmospheric physical parameters is established by using the kernel partial least square regression method, and the obtained model is tested by using the measured spectral data of DR8. The measured results are compared with the atmospheric physical parameters provided by SEGUE SSPP, and good results are obtained. In addition, in order to test the effect of noise on parameter measurement, Gao Si white noise with a signal-to-noise ratio (SNR) of 10 ~ 20 ~ 30 ~ 40 ~ 50 ~ 70 ~ 90120 is added to the Kurucz spectrum, respectively. The Kurucz data with different SNR are tested. The experimental results show that the kernel regression model is sensitive to noise, and the higher the SNR of spectral data is, The higher the prediction accuracy of atmospheric physical parameters is, the less computation and training speed of the kernel partial least square regression model based on line exponents are proposed, which is suitable for the measurement of atmospheric physical parameters of stars.
【作者单位】：山东大学(威海)机电与信息工程学院;
【基金】：国家自然科学基金项目(11078013)资助
【分类号】：P144

【共引文献】