Abstract: Since solar radiation causes the uncertainties and fluctuation of photovoltaic power,which is harmful to the stability and safety of the grid when photovoltaic power generation is connected to the grid,a new super-short-term prediction model of solar radiation whas proposed in this paper. Firstly,a one-dimensional convolution neural network is constructed for data fusion and feature transformation of several key meteorological variables;and then a bidirectional Long Short Term Memory network model is developed to predict global solar radiation in the next 15 minutes. The experimental results show that the proposed model can effectively improve the predicting accuracy compared with the traditional machine learning methods,and that the forecast skill of the proposed model has been improved about 14% over the persistence model on the normalized root mean squared errors.