In order to study the thermal deflection of cable-stayed bridges and obtain the response components of signals, the Nanjing No.3 Yangtze River Bridge was taken as the research object to analyze the multiple time scale characteristics of long-term deflection monitoring signals. The deflection signal was reconstructed on two time scales (diurnal and annual) by using wavelet multi-scale analysis method to realize separation of thermal deflection. The characteristics of diurnal thermal deflection under solar radiation were studied based on long-term monitoring data. Results show that the diurnal thermal deflection had seasonal differences in annual cycle, and it lagged behind the solar radiation in diurnal cycle. There was a strong positive correlation between the diurnal thermal deflection and the solar radiation excluding the lag effect. Considering the influence of season, ambient temperature, and solar radiation, a multi-fold line model of diurnal thermal deflection with prediction error less than 2 cm was proposed, which could predict diurnal thermal deflection at any time in diurnal cycle.