In view of the problem that the lack of solar radiation data leads to the inaccuracy of building energy consumption simulation and building energy conservation analysis, this paper takes Xian as an example to study the selection of typical meteorological year (TMY) and prediction of hourly radiation when the solar radiation data is missing. First, through correlation analysis, it was found that the easily obtained sunshine duration had the highest correlation with solar radiation. Therefore, on the basis of the traditional Sandia method, TMY was selected according to sunshine duration instead of radiation, and the parameters of the selection results were compared and analyzed to verify the accuracy of TMY selected by the new parameters. Then, reasonable prediction input parameters were selected through the comparison of the main influencing parameters of solar radiation, and hourly radiation prediction was carried out by using neural network and its optimization algorithm, which has strong ability in dealing with generalization problems. The obtained results were compared with statistical model and observation data. Finally, an office building model was established according to the building energy conservation design standards in China. The meteorological data obtained by the proposed method were used to simulate and verify the energy consumption of the building, and the changes in the heating and cooling energy consumption were analyzed respectively. Results show that the proposed TMY selection method could well solve the problem of selecting TMY in areas with radiation data missing, and the neural network algorithm could accurately predict hourly radiation data, which provides a new idea for the study of building energy conservation with radiation data missing.