Abstract: In our report, the Elman neural network model was used to predict the hourly global solar radiation in Haikou, and aimed at the shortcomings of the neural network algorithm such as difficulty in the ability of generalization, training approximation, and long training time, a solar total radiation prediction model combined with the gray correlation analysis method was established. Hourly irradiation exposure of global radiation data at the Haikou Weather Station in 2009～2019, and meteorological data such as atmospheric pressure, temperature, relative humidity, precipitation, and sunshine that affect the global radiation were used, the hourly solar radiation prediction model based on Elman neural network with or without similar day was constructed respectively, and for which the data in 2009～2018 were used as the training data set, and the data in 2019 were used as the validation data set, a study of the temporal and spatial distribution characteristics of radiation in Qiongbei area using the spatial interpolation method was performed. The results showed that the correlation coefficient of the global radiation prediction model and the observation value after the similar day screening reaches 0.97, the average relative error is within ±0.5 MJ·m^-2, and the accuracy of this prediction model is better than that of the Elman neural network algorithm. The spatial distribution was even, showing the characteristics of much south and less north. And the temporal distribution of global radiation in Qiongbei area was less in winter and was the highest in summer, and was centered in spring and autumn.