The Application Of Machine Learning In The Research Of Fermiyao Variants

Blazars are the most concerned subclasses of the Active Galactic Nuclei(AGN)The relativistic jets are at a smaller angle to the observer’s line of sight,so it shows some extreme features,such as high-energy γ-ray emissions rapid and large amplitude variability,high luminosity,high and variable polarization,superluminal motions,and very high energy γ-ray emissions.According to the strength of the optical emission line,the blazars are usually divided into BL Lacertae objects(BL Lacs)and flat spectrum radio quasars(FSRQs).After the launch of Fermi Large Area Telescope(Fermi/LAT),thousands of blazars have been detected by Fermi/LAT.And the Imaging Atmospheric Cherenkov Telescopes(IACT)like HESS only found 73 blazars with TeV emissions,and 71 of them have been detected by Fermi/LAT.Obviously,it is necessary to enlarge the sample size for studying the very high energy γ-ray emissions of the blazars.Besides,there are many blazars with optical classification uncertainty(BZUs).They have different names in different sources catalogs,which are called BCUs in the Fermi sources catalogs,while BZUs in 5BZCAT catalogs.To evaluate the optical classification of BCUs/BZUs is of scientific significance for understanding the classification of blazars.Therefore,in this paper,we will use machine learning method to do two aspects of work,one is to find TeV blazars candidates from Fermi blazars,the other is to evaluate the optical classification of some BZUs.The structure of this paper is as follows,Chapter 1 introduces the observation and classification of AGN,the high energy emissions of blazars and the research status of BCUs/BZUs;Chapter 2 introduces supervised machine learning methods;Chapter 3 and Chapter 4 describe our work as followsIn chapter 3,we collect 418 Fermi blazars in 3LAC clean sample and get 12 pa-rameters,such as redshift and multi-band observation data.Using the feature selec-tion method in supervised machine learning,we get the 3 most important parameters to distinguish TeV and non-TeV blazars:log fX,log fγ1 and log dL.Then,by using logistic regression(LR)classifier,we find the classification criteria of TeV and non-TeV blazars in 3-parameter space,and apply it to the 418 Fermi blazars,thus we get 35 TeV blazar candidates.An empirical discriminant formula is obtained:logit=2.753 log fx+2.582 log fγ1-2.714 log dL+179.8,if logit>0,the probability that the source is a TeV blazar candidate is more than 50%.We also tried to fit the energy spectra(SEDS)of 35 Tev candidates.And it is found that 12 of the 35 candidates can be detected by LHAASOIn Chapter 4,we collect 1425 BL Lacs and 1909 FSRQs from 5BZCAT and get 8 pa-rameters,such as redshift and multi-band observation data.Feature selection and feature extraction are used to find the optimal parameter space.Then,four supervised machine learning classifiers are used to find the classification criteria of BL lacs and FSRQs in the optimal parameter space.and apply it to the 227 BZUs.Then 33 BL lacs candidates and 119 FSRQs candidates are obtained,and 75 BZUs are not successfully classified.