Detection of solar flare effects in the amplitude variations of VLF radio signals using an autoencoder

Detection of anomalies in changes in radio signal parameters is one of the tools for studying the dynamic properties of a radio communication channel. The development of machine learning technologies allows us to take a new approach to solving pressing problems in radio physics. The purpose of this work is to develop an algorithm for searching for the effects of solar flares in variations in the amplitude of VLF radio signals using machine learning methods. The developed models of a neural network based on an autoencoder are presented for automatic detection of anomalous amplitude changes based on the propagation data of VLF radio waves on the JJI (Japan, 22.2 kHz) – Irkutsk (receiving point in Teploenergetik) path. After training, the neural network reconstructs the daily variation of the signal amplitude so that the output data closely matches the propagation under quiet conditions. Anomalous signal changes associated with ionospheric disturbances during solar flares are detected by determining the reconstruction error and comparing it with a calculated threshold value. The threshold is determined by calculating the root mean square value of the reconstruction error of the training data set. The results of using the autoencoders proposed by the authors to search for anomalies have shown their effectiveness (the overall accuracy was close to 90%). An important advantage of the method is the ability to perform an assessment using data only for the measurement day in question, which eliminates the influence of anomalies present on neighboring days.