UDC 004.032.26; 621.371
FORECASTING OF DAILY AMPLITUDE VARIATIONS OF VLF RADIO SIGNALS USING RECURRENT NEURAL NETWORKS
Nguyen Khac Hoang Duong, post-graduate student, INRTU, Irkutsk, Russia;
orcid.org/0009-0000-1561-8741, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
A. S. Poletaev, Ph.D. (Phys. and Math.), Associate Professor, INRTU, Irkutsk, Russia;
orcid.org/0000-0002-6448-0045, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
A. G. Chensky, Ph.D. (Phys. and Math.), Associate Professor, INRTU, Irkutsk, Russia;
orcid.org/0000-0003-2076-5396, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
The article explores the application of Long Short-Term Memory (LSTM) recurrent neural networks to forecast the daily variations in the amplitude of Very Low Frequency (VLF) signals. Based on the data collected at the receiving station in Irkutsk along JJI (Japan) – Irkutsk transmission path with a frequency of 22.2 kHz, a model to predict daily amplitude variations of VLF signals over various time intervals has been developed. The data from July 1 to July 16, 2017, were used for training and testing the model, with prior preprocessing to optimize computational resources. Experimental studies confirmed that the model achieves high accuracy in long-term forecasting. Specifically, for a 24-hour forecast, average error values were as follows: RMSE – 0.000596, MAPE – 11.51 %, and NMAE – 5.96 %. The analysis of the results demonstrated that the model effectively captures the overall trend of daily amplitude changes, with higher prediction accuracy for signals that are closer in season and conditions to training data. These findings validate the effectiveness of LSTM neural networks in predicting VLF signal amplitudes and highlight their potential for applications in radio communication and ionospheric monitoring.
Key words: : : VLF signal, ionosphere, machine learning, neural network, time series, RNN, LSTM, forecasting.