In view of the shortcomings that the accuracy of the existing ionospheric models is not enough to reflect the local real-time changes, this paper uses the time series prediction tool of deep learning to obtain the model of a specific communication link, and combines it with the ionospheric theoretical calculation model to improve the accuracy of communication link prediction. The main work is to analyze and compare the international broadcasting data and the model calculation data, summarize and sort out the characteristics of the link data, and then on this basis, look at the change of the difference between the two, and carry out link prediction, which achieves the goal of improving the calculation effect of the link model, which is of great significance to the short-wave frequency management.