Research On Automatic Measurement Of Stellar Atmospheric Parameters Based On KPCA And Elastic Net Regression

Most human cognitions about the nature of stars are derived from stellar spectra.The spectrum contains a wealth of information.With the sixth-period spectral observation task of the Large Sky Area Multi-Object Fibre Spectroscopic Telescope(LMAOST)had been completed.Therefore,it is especially important to find an effective and accurate method for automatic processing of massive stellar spectra.This article is developed in this context,focus on the automatic measurement of atmospheric parameters of stellar spectra,meeting the needs of the LAMOST project.This article uses LAMOST DR6 astronomical spectrum data,to study the application of elastic net regression(ENT)method in the automatic measurement of stellar atmospheric parameters.The main research work of this paper is as follows:1?Preprocessing and feature extraction of high dimensional spectral data.Feature extraction plays a key role in the measurement of celestial spectrum parameters.This paper chooses to extract the mathematical features of the spectrum,first preprocesses the spectrum,and then uses the principal component analysis to extract the features of the spectral data.2?Application of elastic net regression algorithm in stellar atmospheric parameter measurement.This section first introduces the theory of elastic regression,then uses PCA to preprocess the measured spectral data of stars,and then uses the elastic regression method to measure the physical parameters of the star's atmosphere.The experimental comparison results show that the elastic regression algorithm is feasible and good in predicting the atmospheric parameters of stars.The ENT model's prediction accuracy for stellar atmospheric parameters is better than that of the SVR model.3?Elastic net regression stellar atmospheric parameter measurement based on kernel principal component analysis.In this chapter,we first carry out comparative experiments on the combination of KPCA(kernel principal component analysis),FA(factor analysis),and the combined PCA-FA,KPCA-FA,and elastic regression,study the impact of different extraction methods on parameter measurement,the results show that the prediction accuracy is higher when using kernel principal component analysis for feature extraction.Then,the kernel function selection experiment and the kernel parameter selection experiment are carried out,aiming at optimizing the elastic regression model to make it have better prediction performance.Finally,in order to verify the predictive ability of the model,a stellar subclass parameter measurement experiment and a different method comparison experiment were performed.The above-mentioned overall experiments show that the KPCA-ENT based stellar atmospheric parameter measurement method is feasible,and good prediction results can be obtained for stars and A,F,G,and K stars,and classification prediction will make each type of star The prediction accuracy has been improved.4?Multi-task elastic net regression stellar atmospheric parameter measurement based on kernel principal component analysis.The model truly realizes the simultaneous modeling and prediction of three physical parameters.Compared with the single elastic regression model,the prediction efficiency can be improved by 2 times and the prediction accuracy is still good.At the same time,the model is applied to the measurement of stellar subclass parameters.The experimental results show that the model has high prediction accuracy for the measurement of stellar and A,F,G and K stellar parameters.