Automatic Measurement Of Celestial Spectral Parameters Based On Kernel Ridge Regression Method

There are abundant physical and chemical information in celestial spectrum.The research of celestial spectrum data is mainly divided into quantitative analysis and qualitative analysis.Qualitative analysis mainly refers to the determination of the chemical composition of celestial bodies,while quantitative analysis is to determine the content of chemical elements in celestial bodies,such as temperature,pressure and other parameters,and then determine the relevant scientific attributes of celestial bodies indirectly.Currently,Large sky Area Multi-Object Fiber Spectroscopic Telescope(LAMOST,also known as Guo Shoujing Telescope)is the highest spectral acquisition telescope in the world,its establishment and operation are of great significance for astronomical research,and it has a far-reaching impact on the development of astronomical industry of China.In this paper,the use of LAMOST celestial spectrum data,the study of nuclear Ridge Regression(KRR)application of the algorithm in terms of measuring physical parameters of the atmosphere of the star.The main contents are as follows:(1)Preprocessing and feature extraction of original spectral data.Spectral preprocessing usually includes spectral de-noising,flux normalization,continuous spectrum fitting,spectral line extraction,and feature extraction includes physical feature detection and mathematical feature extraction.A one-dimensional spectral data usually corresponds to a vector of thousands of components.Feature extraction plays a key role on the measurement of scientific parameters and the inference of properties of celestial objects.In this paper,principal component analysis is used to extract features from spectral data.This method can preserve the main features of data and reduce the storage of data and the complexity of calculation at the same time.(2)Automatic measurement of stellar atmospheric parameters based on kernel ridge regression algorithm.Firstly,this paper introduces the theoretical development of the ridge regression algorithm and the kernel function.Ridge regression method is a distortion of least squares method,error terms are added on the basis of least squares method.The unbiased estimation problem is transformed into biased estimation,which enlarges its application scope with losing part of accuracy.Secondly,the role and selection of the kernel function are studied.The combination of the kernel function and the ridge regression method is to obtain the kernel ridge regression(KRR)algorithm.Based on this method,the automatic measurement of stellar atmospheric physical parameters is carried out.The experimental results and error analysis show that the method can automatically measure the stellar atmospheric parameters with high accuracy.(3)Automatic measurement of stellar atmospheric parameters based on data grouping strategy plus kernel ridge regression algorithm.Compared with other algorithms,such as support vector machine regression(SVR),kernel ridge regression has a disadvantage of long running time,but the measurement results are slightly better.So,In order to improve the efficiency of the KRR method,this paper combines the data clustering the KRR method to reduce the running time.This paper compares random clustering and K-means clustering analysis based grouping,and finds that K-means method is more effective.In addition,due to the improvement of data utilization efficiency,the running time is reduced,and the accuracy of parameter measurement results in the experiment is also improved accordingly.