Optimised CNN-LSTM-Attention deep learning network model based on RIME algorithm for predicting HFSWR gravity wave parameters during Typhoon Muifa (2022)

The gravity waves as a medium can convey energy during typhoon operation from the troposphere to the ionosphere, which have an essential influence on the ionosphere and atmospheric structure. High-frequency surface wave radar (HFSWR) serves as an efficient tool for detecting gravity waves. Through advanced data processing, the study extracts gravity wave power data from the HFSWR observations made during a typhoon. This research proposes a RIME algorithm optimised CNN-LSTM-Attention network model (RCLAN) to achieve effective prediction of gravity wave power data. The RIME optimization algorithm is utilized to enhance the model's ability to extract features, and simultaneously, the integration of attention mechanisms within the regression prediction component is focused on elevating the precision of predictions. In addition, in the research, the prediction results of the proposed RCLAN gravity wave power forecasting model are compared with the prediction results of the traditional convolutional-long short-term memory model (CLSTM), long short-term memory network model (LSTM), and other models and the error values are calculated. With the data in data set (a), for instance, the RMSE (root-mean-square error) value of the RCLAN model is 0.0171, which is 42.75% smaller than the lowest RMSE value of the rest of the models, and the error of the RCLAN model in the rest of the data groups is also at a low level indicating that the proposed RCLAN model has better prediction accuracy and robustness. This study proposes the RCLAN model to achieve accurate prediction of gravity wave power data, and the obtained predictions can provide research material for future studies on the relationship between gravity waves and typhoons.