A BWO-VMD-CNN-BiLSTM-Attention Model for HF Communication foF2 Prediction in the Chinese Middle Latitude Region

The critical frequency in the F2 layer (foF2) is an important parameter in the HF communication, for the reason that F2 layer can reflect shortwave radio waves. In this paper, we consider using the optimized signal analysis method combined with deep neural network to predict the foF2 sequence changes of nine links in the middle latitude region of China. The combined model used in this paper is composed of Variational Mode Decomposition (VMD) under the Beluga Whale Optimization (BWO) algorithm, Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM) neural network, and attention mechanism layer. Variational mode decomposition is used to reduce the impact of non-stationary features on prediction, and CNN and BiLSTM combine with the attention mechanism layer to extract the change features in the foF2 sequence. By using the historical ionospheric information of 9 links to train the implemented model, that is BWO-VMD-CNN-BiLSTM-Attention model, the results show that the implemented model reduces the prediction error (RMSE) by 21.72 % and 23.71 % respectively compared with the CNN-BiLSTM-Attention model and the BiLSTM model, and the prediction performance is improved under different detection links.