Prediction of Marine Atmospheric Duct and its Loss Based on LSTM

The most commonly used model for atmospheric data prediction is the WRF model, which is based on physical laws and simulates atmospheric movement and changes by solving processes in atmospheric dynamics, thermodynamics, and radiative transfer. However, the high demand of the WRF model for computing resources and time presents challenges for computer performance. This study aims to enhance the computational speed and generalization ability of the model by incorporating LSTM recurrent neural networks for predicting future atmospheric data based on historical records. Compared to traditional statistical and physical models, LSTM demonstrates superior adaptability in analyzing complex atmospheric data patterns and rules, thus achieving better performance and applicability in atmospheric data prediction. We used the WRF software to generate a dataset comprising air pressure, temperature, water vapor pressure, wind speed, and wind direction within a 10 km altitude in the South China Sea region. The neural network model is trained to predict air pressure, temperature and water vapor pressure, and to retrieve the ocean evaporation layer. The results show that the mean square error (MSE) of LSTM predicted meteorological data is lower than that of WRF. Using the predicted data, the transmission loss of electromagnetic wave is calculated by using the parabolic equation method, and the prediction of electromagnetic wave transmission loss is realized. Using the data generated by the prediction, the parabolic equation method (PE) is used to calculate the transmission loss of electromagnetic wave, and the prediction of the transmission loss of electromagnetic wave is realized.