The reliability of the measurement of ultraviolet radiation has always been a hot spot of research. The observation of ultraviolet radiation is not only affected by the solar elevation angle, aerosol thickness, ozone, dioxide, there is also a great connection with the systematic error of the measuring instrument. In fact, in the ultraviolet radiation observation, due to the lack of routine maintenance and periodic calibration, the radiation meter will obviously decline after a period of time, and the longer the use time, the more obvious the attenuation. Therefore, in order to obtained the consistent time series of the stable observational values, some reasonable methods must be adopted to correct the measured values. The data source of this research was part of the UV-MFRSR type ultraviolet radiometer observations from 2003 to 2010. These data were obtained by these daily time series calibration method. In theory, these time series points represent the response time of the instrument, and they should be stable for several months or even years. However, the performance of the in-situ calibration method was influenced by the aerosol / ozone loading mode in practice. The purpose of this study was to get a smooth observation curve by eliminating some observational anomalies. In addition, the actual data in the observation process, some date data is missing, so the reasonable prediction model is used to estimate the value of these data. In this paper, the ARIMA and GARCH models were used to predict the missing data and compared between the predicted value and the true value, it is found that the fitting degree of the predicted value and the true value based on the AR-GARCH model is higher.