Abstract: The hourly solar radiation data of Beijing and Wuhan from 2001 to 2015 are collected,and the typical meteorological years are selected for the two places. According to the data,the influence of the diffuse fraction of the two places on the meteorological environment factors is discussed. Under the circumstance of cross-subdivision of weather types,the appropriate mathematical models for each place are established and compared.The results show that:1)the diffuse fraction in Beijing is more strongly correlated with the clearness index,and the linearization degree is better. 2)the influence of solar altitude angle on the diffuse fraction in different regions is different,and the difference is obvious;3)under the weather types with more meteorological impact factors,the LMBP neural network model has a strong applicability in different regions. When the meteorological impact factors are less,the linear model in Beijing is better than the improved model,while the linear model in Wuhan is less effective than the improved model in some weather types.