For citation:
Appalonov A. M., Maslennikova J. S. A hybrid total electron content forecasting model based on autoencoders and classical machine learning algorithms. Izvestiya VUZ. Applied Nonlinear Dynamics, 2026, vol. 34, iss. 2, pp. 286-298. DOI: 10.18500/0869-6632-003207, EDN: QKRVJO
A hybrid total electron content forecasting model based on autoencoders and classical machine learning algorithms
Purpose. Development of a novel two-stage machine learning algorithm for total electron content (TEC) forecasting based on original TEC time series and influential external ionospheric parameters.
Methods. Dimensionality reduction of the input data is performed using a standard fully-connected autoencoder to obtain compressed latent representations. These features are integrated with a set of external parameters: the critical frequency of the F2 layer (foF2), solar (F10.7) and geomagnetic (Kp) activity indices, and temporal descriptors (seasonal and diurnal information). The enriched dataset is used to train and evaluate several classical machine learning algorithms, including gradient boosting (CatBoost), with assessment
based on RMSE and MAE metrics.
Results. The CatBoost algorithm demonstrates superior predictive accuracy on the test dataset compared to other evaluated models. The proposed two-stage approach proves effective for extracting and utilizing key temporal dependencies for the regression task.
Conclusion. The developed method provides accurate TEC prediction by combining neural network-based time series compression with modern ensemble algorithms, as confirmed by the computational experiment.
- Brjunelli BE, Namgaladze AA. Physics of the Ionosphere. M.: Nauka; 1988. 527 p. (in Russian).
- Mendillo M. Storms in the ionosphere: Patterns and processes for total electron content. Rev. Geophys. 2006;44(4):RG4001. DOI: 10.1029/2005RG000193.
- Maksimov DS, Kogogin DA, Nasyrov IA, Zagretdinov RV. Solar flares in the 25th cycle of activity: Effect on ionospheric disturbance and GNSS signal strength. Current Problems of Remote Sensing of the Earth from Space. 2025;22(3):301–317 (in Russian). DOI: 10.21046/2070-7401-2025-22-3-301-317.
- Fitzgerald TJ. Observations of total electron content perturbation of GPS signals caused by a ground level explosion. Journal of Atmospheric and Solar-Terrestrial Physics. 1997;59(7):829–834. DOI: 10.1016/s1364-6826(96)00105-8.
- Feng JD, Zhang T, Li W, Zhao Zh, Han B, Wang K. A new global TEC empirical model based on fusing multi-source data. GPS Solutions. 2023;27(1):20. DOI: 10.1007/s10291-022-01355-8.
- Bilitza D., Pezzopane M., Truhlik V., Altadill D., Reinisch B.W., Pignalberi A. The international reference ionosphere model: A review and description of an ionospheric benchmark. Rev. Geophys. 2022;60(4):e2022RG000792. DOI: 10.1029/2022RG000792.
- Natras R, Soja B, Schmidt M. Ensemble machine learning of Random Forest, AdaBoost and XGBoost for Vertical Total Electron Content forecasting. Remote Sens. 2022;14(15):3547. DOI: 10.3390/rs14153547.
- Appalonov AM, Maslennokova YuS, Sherstyukov ON. Application of deep learning neural networks for the analysis of spatial and temporal components of the decomposition of the total elec-tronic content of the ionosphere. Radioengineering. 2025;89(1):172–179 (in Russian). DOI: 10.18127/j00338486-202501-16.
- Appalonov AM, Maslennikova JuS, Sherstyukov ON. Analysis of spatiotemporal variations of the complete electronic content and the critical frequency of the F2 layer using deep learning neural networks. Propagation of radio waves [Electronic resource]. In: Collection of Reports of the XXIX All-Russian Open Scientific Conference. June 30–July 4, 2025, Kazan, Russia. Kazan: Kazan University Publishing, 2025. P. 512–515. Available from: https://repository.kpfu.ru/?p_id=319113.
- JPL. Official site of JPL [Electronic resource]. Available from: https://www.jpl.nasa.gov.
- Global Ionospheric Radio Observatory. Official site[Electronic resource]. Available from:linebreak https://giro.uml.edu/.
- Dominion Radio Astrophysical Observatory (DRAO) : official site / National Research Council Canada [Electronic resource]. Available from: https://www.cadc-ccda.hia-iha.nrc-cnrc.gc.ca.
- Helmholtz-Zentrum Potsdam – Deutsches GeoForschungsZentrum (GFZ). Official site / Helmholtz Association[Electronic resource]. Available from: https://www.gfz.de/.
- Appalonov AM, Maslennikova JuS, Sherstyukov ON. Decomposition of global maps of the complete electronic content of the ionosphere using neural networks. In: Proceedingslinebreak of the 22nd International Conference ``Modern Problems of Remote Sensing of the Earth from Space''. М.: Russian Space Research Institute Publishing; 2024. P. 434 (in Russian). DOI: 10.21046/22DZZconf-2024a.
- Appalonov AM, Maslennikova YuS. Short-term prediction of the total electronic content of the ionosphere using solar parameters by machine learning methods. In: Proceedingslinebreak of the 21 International Conference ``Modern Problems of Remote Sensing of the Earth from Space''. М.: Russian Space Research Institute Publishing; 2023. P. 299 (in Russian). DOI: 10.21046/21DZZconf-2023a.
- Masini RP, Medeiros MC, Mendes EF. Machine learning advances for time series forecasting. Journal of Economic Surveys. 2021;37(1):76–111. DOI: 10.1111/joes.12429.
- 457 reads