V.A. Aporovich1, P.S. Sushko2

1, 2 OJSC "AGAT-Control Systems" – Managing Company of "Geoinformation Control Systems" Holding (Minsk, Belarus)

1 aporovich@agat.by, 2sushko-ps@agat.by

The article proposes a new filter for smoothing the coordinates of a maneuvering aerodynamic target using neural networks.

The purpose of the work is to explore the possibility of using neural networks to solve this problem. It is proposed to use recurrent neural networks as a neural network architecture. This type of architecture allows you to take into account the past states of the target explicitly, as a sequence over time, which is much more efficient than using classical fully connected direct distribution networks. The classical RNN algorithm faces the problem of decaying gradients, which significantly impairs the quality of neural network training. The most effective solutions are to use LSTM or GRU instead of the classic RNN. To train the neural network, data was generated, which is a set of fixed-length sequences, where the target variable is the difference between the true coordinate and the coordinate with an error. The effectiveness of the filter using RNN was evaluated using simulation modeling. The simulation results of the RNN filter were compared with the results of other filters.

The simulation showed that in terms of the root-mean-square error of the smoothed coordinate, RNN occupies an intermediate position between multi-alternative filters (IMM) and an effective filter with coefficient variation.

Thus, the proposed filter has shown its efficiency and the possibility of further improvement to solve problems in radar information processing systems.

Aporovich V.A., Sushko P.S. Smoothing the coordinates of a maneuvering aerodynamic target using a neural network // Information-measuring and Control Systems. 2026. V. 24. № 4. P. 14−19. DOI: https://doi.org/10.18127/j20700814-202604-02

1. Khmarskii P.A. Obrabotka rezultatov izmerenii koordinat i parametrov dvizheniya v sistemakh monitoringa vozdushnoi i nazemnoi obstanovki: monografiya. Pod red. V.M. Artemeva. Natsionalnaya Akademiya nauk Belarusi. In-t prikladnoi fiziki. Minsk: Belaruskaya navuka. 2025. 261 s. (in Russian)
2. Aporovich V.A. Sintez i issledovanie filtra s variatsiei koeffitsientov dlya soprovozhdeniya manevriruyushchei aerodinamicheskoi tseli. Informatsionno-izmeritelnye i upravlyayushchie sistemy. 2021. T. 19. № 6. S. 21−28. DOI: https://doi.org/10.18127/j20700814-202106-02. (in Russian)
3. Yang Chun, Blasch Eric. Characteristic Errors of the IMM Algorithm under Three Maneuver Models for an Accelerating Target. 11th International conference of information Fusion. IEEE. Cologne. Germany. 2008.
4. Blackman S., Popoli R. Design and Analysis of Modern Tracking Systems. Artech House. Boston. London. 1999.
5. Schmidt R.M. Recurrent Neural Networks (RNNs): A gentle Introduction and Overview. arXiv preprint arXiv: 1912.05911. 2019.
6. Chung J., Gulcehre C., Cho K., Bengio Y. Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling. arXiv preprint arXiv: 1412.3555. 2014.
7. Guizhu Shen, Qingping Tan, Haoyu Zhang, Ping Zeng, Jianjun Xu. Deep Learning with Gated Recurrent Unit Networks for Financial Sequence Predictions. 8th International Congress of Information and Communication Technology. ICICT. 2018. P. 895−903.
8. Hyndam R.J.,Athanasopoulos G. Forecasting: principles and practice. 2018. Edition 2nd. OTexts: Melbourne. Australia. OTexts.com/fpp2/residuals.html (04.04.2026).
9. Heato J. An Empirical Analysis of Feature Engineering for Predictive Modeling. arXiv prepring arXiv: 1701.07852. 2020.