| The search and research of special and rare objects like cataclysmic variables has always played an important role in astronomy.Cataclysmic variable is one kind of binary system consisting of a white dwarf and a companion star which is normally a red giant stars filled with Roche lobes or a late main-sequence star.It is also a class of celestial bodies with abundant members and many types of variable stars in the field of binary stars and variable stars in astrophysics.Cataclysmic variable can be observed and studied from radio telescopes,infrared radiation to X-ray wavelengths.The spectrum of cataclysmic variable is the basis of studying its physical parameters.However,relevant research on cataclysmic variable has been greatly restricted due to the limited amount of observational spectra.Of all the confirmed cataclysmic variables stars,only 457 have spectroscopic observations and most of them are observed and confirmed through SDSS spectra.Therefore,it is of great necessity to expand the observational spectra library of cataclysmic variables.At present,the research of astronomical spectrum has entered an era of big data,Large Sky Area Multi-Object Fiber Spectroscopic Telescope(LAMOST)has the highest spectral acquisition rate in the world.After LAMOST completes the sky survey,its spectral number will be 10 times that of Sloan Digital Sky Survey(SDSS),which makes it possible for us to search for cataclysmic variables stars in a wider sky area.It is certain that LAMOST will increase the number of cataclysmic variables by an order of magnitude.The project mainly consists of two parts.Firstly,the machine learning methods are studied and applied to automatic search of cataclysmic variables from the massive LAMOST sky survey spectroscopic data.After the sample has been enlarged,the parameters of the cataclysmic variables are measured based on the photoionization model.The specific work is as follows:(1)The spectra of LAMOST and SDSS were used as multi-source template data.The template spectra by improved AAE network and the spectra of the certified cataclysmic variables were then mixed as template spectra.Models then were classified by the LightGBM algorithm based on integrated tree models,and this classifier was used to search for cataclysmic variables candidates in LAMOST-DR7 automatically.After being verified by Simbad and the cataclysmic variables lists,10 unrecorded cataclysmic variables candidates were successfully identified,including one with absorption characteristics,which proves the feasibility of the algorithm.Meanwhile,models of GBDT,Random forest,XGBoost and LightGBM based on the ensemble tree model were built on three different data sets,and the four models were compared comprehensively using every indicator.The results indicated that the indicators of LightGBM were better than the other three models,and were more suitable for processing spectral data of large-scale and high dimension,which also proved the validity of IAAE.The LightGBM model gave the importance score of each wavelength.By analyzing these importance scores,it was found that the features extracted by the model were consistent with the actual features of the spectra,which further proved the accuracy and superiority of the model.(2)The parameters of the cataclysmic variables were measured by CLOUDY,which is a spectral simulation software based on photoionization.By inputting a few physical parameters,it simulated the emission spectra of cataclysmic variables in the quiet period and cross matched with the measured spectra.The comparison results showed that there was a high similarity between the two in spectral type and peak value of particular wavelengths. |
