Artificial Neural Network For Constructing Type Ia Supernovae Spectrum Evolution Model

About two decades ago,the accelerated expansion of the universe was found through the observation of type Ia supernovae(SNIa).To explain this phenomenon various cosmological models were put forward.Some type Ia supernovae survey projects with different designs were carried out subsequently in order to constraint cosmological model parameters.Absolute magnitude of type Ia supernovae are almost same after corrected its brightness,so it is very important to measure the distance of type Ia supernovae accurately as standard candlelight to limit the parameters of cosmological model.Therefore,some methods and supernova models have been invented to measure the distance of IA type supernovae accurately.Some of these techniques need too much spectrum to cover the supernova parameter space,This leads to new errors in the final results.Currently,obtaining the parameters of supernova data set used to fit the cosmological model by fitting the observed data with the SALT2 model.But the model assumes some relationships between SNIa's spectral and its parameters,and the model parameters are more than 3000.With the development of technology,this paper attempts to use artificial neural network for constructing SNIa spectrum evolution model.The parameters,this methods needed,are significantly less than the SALT2 model(about one third of the parameters of SALT2).And the artificial intelligence technology does not need to assume any relationship between SNIa's pa-rameters and its spectral energy distribution,but let algorithms automatically find the relationships between inputs and outputs during the training process.Because the neural network has a good ability of fault tolerance,even if there are some bad data,final results will not be affected.The model can generates spectra of different SNIa that trained by using the data of supernovae spectra and light-curve.Conse-quently,the model can simulate SN's spectra during supernova explosion process,obtaining light-curve by simulated the telescope to observe it at that time.The results,given by the network,are in good agreement with the observation results.Accordingly,this model can also be used to measure the distance of SNIa accurately and study the spectral differences of different supernovae.The net is appropriate for research derivative of supernova spectrum to its parameters duo to it is differen-tiable.The first chapter introduces the research status and significance briefly.In the second chapter,introducing the mechanism of supernova explosion and observation of supernovae briefly,and then the spectra data and the light curve data as well as the preprocessing of these data are introduced.The neural network model and the algorithm used to train it are depicted in Chapter 3.Computer programs of neu-ral network and corresponding algorithm are written;Because the current popular framework of artificial neural network is not suitable for solving our problems.In order to study the average spectrum of type Ia supernovae,the relationship between SN's spectrum and its parameters and to determine the parameters.Networks of different structure are trained for different purposes,then comparing the results of these nets.In the fifth chapter,we use neural network to generate spectra of different supernovae,and compare these spectra similarities and differences.Then studying how evolution the speed of the projectile of SNIa with phase.Finally,given cosmo-logical constraints from a fitting of the SNIa's data set,cosmic microwave background data and baryon acoustic oscillation data.