Light Curve Modeling of Semi-detached Binaries Based on Neural Network

Semi-detached binaries are significant targets for the study of the formation and evolution of interacting binaries. Rapid modeling tool is highly required to derive the parameters with large amount of stars to be observed by many recent time-domain photometric surveys. In this work, based on a neural network, a light curve modeling of semi-detached binaries is proposed, which can derive orbital inclination (incl), relative radius (R/a), the mass ratio (q), and temperature ratio (T₂/T₁) fast via the observational light curve and known effective temperature of the primary star. The results of Kepler's light curve modeling show that the model can accurately fit the light curves of pulsating eclipsing binaries (the fitting degree can reach more than 0.9). For a target whose relative measurement error, orbital inclination, the amplitude of light curve, and temperature ratio are 0.01, ~90°, 1.84 mag, and 0.6, the measurement errors are 1.251, 0.004, 0.008 and 0.003 for incl, R/a, q, and T₂/T₁, respectively. In addition, as an application, the proposed model in this work can be deployed on other photometric data by simply replacing the train data, which provides an effective tool to obtain a large number of parameters of semi-detached binaries and fast search for candidates of abnormal binaries.