Aimed at the problems that in the existing high frequency communication, the frequency parameter prediction methods are tedious formalities in operation and shortage in precision, this paper presents a prediction model of frequency parameters of shortwave communication based on long shortterm memory recurrent neural networks. This neural network can break through the limitations of traditional neural networks and establish longterm correlations on data sequences. The experimental results show that the mean square error (MSE) can be control below 2% and the model reduced the error by 7%. And this method is effective and superior to the traditional prediction method.