Research On The Application Of Multi-learner Algorithm In The Measurement Of Stellar Atmospheric Parameters

Each phase of LAMOST survey will produce a large amount of celestial spectral data.In order to use these spectral data for the study of the Milky way,we first need to extract the relevant physical information from these spectral data.Therefore,combined with the background of astrophysics and through mathematical theory and computer methods,the study of automatic measurement of stellar physical parameters can not only improve the efficiency of the use of LAMOST data,but also find the physical characteristics and laws in spectral data.Based on this,this paper mainly studies the application of multi-learner algorithm in the automatic measurement of stellar atmospheric parameters by using LAMOST celestial spectral data.The main research work of this paper is as follows:(1)Feature extraction of celestial spectral data line index.Before the parameter measurement of celestial spectral data,it is necessary to extract the relevant features of the spectrum.In the past,most methods such as PCA are used to extract the mathematical features of the spectrum,but the effective features extracted can not be specifically known.In this paper,the line index,which is the effective physical feature of celestial spectrum is extracted.The use of line index not only effectively reduces the dimension of spectral data,but also defines the detailed physical features of the spectrum.(2)Research on measurement of stellar atmospheric parameters based on weak learner lifting.In this paper,firstly,the spectral data of type A stars are processed,and then the algorithms of Adaboost,GBDT,XGBoost and Light GBM multi-learner are theoretically analyzed and studied.Finally,the stellar atmosphere parameters are measured based on the four algorithms,and the prediction results are compared with those of the classical algorithm SVR.The experimental results show that the Light GBM algorithm can be well applied to the measurement of stellar atmospheric parameters,and based on the A-star spectral data,the prediction effect of Light GBM algorithm is the best.(3)Measurement of stellar atmospheric parameters based on multi-model fusion.The prediction effect of multi-learner algorithm is excellent,but there is still room for optimization.This paper first makes a theoretical analysis of Stacking multi-model fusion,and then carries out different combinations of multi-model fusion experiments based on the above four multi-learner algorithms to find the optimal multi-model combination scheme.The experimental results show that the multi-model fusion is carried out on the basis of multi-learner algorithm,the first layer uses Adaboost,GBDT and XGBoost algorithm,and the second layer uses Light GBM algorithm to get the best model combination effect.Finally,the stellar atmosphere parameters are automatically measured based on the optimal combination of multi-model fusion,and the results show that the optimization of multi-model fusion to multi-learner algorithm is feasible,and the prediction accuracy of stellar atmospheric parameters has been improved in varying degrees.