Study of Applicability of LSTM Neural Networks for Predicting the Characteristics of Short-Wave Communication Channels

The developed method of short-term prediction of the maximum usable frequency (MUF) characteristic of shortwave (SW) communication channels is investigated on the basis of a neural network combining a combination of layers with long short-term memory and full-link layers. The synthesis of the structure is performed using the Keras tuner framework. The validation of the methodology was performed on the basis of the maximum usable frequency data of a 2.6 thousand km midlatitude trace. The accuracy of the predictions was evaluated using R2, MAPE, MAE and RMSE metrics.