Neural Network Model of the Effective Temperature for A Type Star Based on Principal Component Analysis

The Large Sky Area Multi-Object Fiber Spectroscopic Telescope (LAMOST) has provided bulk of stellar spectra data. DR5 catalogue contains plenty of spectral indices and effective temperature of A type stars. Recently machine learning algorithms such as Neural network model which can be used to explore the deep relationship between different data have been widely used in various disciplines. In this paper with 19 spectral line indices and effective temperature of A type star from LAMOST DR5 data set are involved. Through Principal Component Analysis (PCA), we present the percentage of the entire information for each spectral index and 12 spectral line indices which are most closely related to effective temperature are selected as an input to establish a Neural network model for effective temperature, meanwhile the absolute error of effective temperature for these input data are less than 100 K. The model performs well overall on the test data set. The coefficient of determination R² given by the program is 0.904 and an average absolute error of 58.38 K. Compared with related research model, the measuring accuracy has been significantly improved. Furthermore, for the raw data which have absolute error more than 100 K, we remeasure effective temperature via our model and the average absolute error of the new effective temperature data has decreased significantly. Besides, LAMOST DR5 catalogue barely includes effective temperature of A5 type star, we replenish these missing data. This work provides a certain degree of reference significance.