Analysis of parameters of basic neural network architectures in transport control systems
https://doi.org/10.46973/0201-727X_2026_1_20
Abstract
The article considers an approach to improving the efficiency of automated train control systems by integrating a neural network assistant module into the contour of the automatic blocking logical control (ABLC) subsystem. An analysis of the applicability of various neural network architectures for solving problems of intelligent support for train dispatchers is carried out, including multilayer perceptron (MLP) and long short-term memory networks (LSTM). Based on a simulation model of a dispatching section comprising three stations, quantitative assessment of the dimension of the input vector of the state a quantitative assessment of the input state vector dimension and computational complexity of the considered architectures is performed. Formulas for calculating the number of trainable parameters are presented, and a comparison of MLP and LSTM with identical model parameters is made. The results of the comparative analysis demonstrate the advantage of LSTM in terms of computational efficiency while preserving the structure of temporal dependencies. Based on the obtained results, the choice of LSTM as the basic architecture for building an intelligent assistant operating in a single contour with the ABLC subsystem is substantiated.
About the Author
I. O. Alymov
Rostov State Transport University (RSTU)
Russian Federation
Russian Federation
Alymov Ilya Olegovich, Chair “Railway Automation and Telemechanics”, Postgraduate Student
References
1. Alymov, I. O. Problems of safety and reliability in automated train traffic control systems / I. O. Alymov // Transport : Science, Education, Production : collection of scientific papers of the International scientific and practical conference, Rostovon-Don, April 23–25, 2025. – Rostov-on-Don : Rostov State Transport University, 2025. – P. 28–32. – EDN CULUWE.
2. Features of the emotional state of railway transport workers / N. N. Malyutina, A. L. Sedinin, S. V. Luzina, and N. S. Sedinin // Journal of scientific articles “Health and Education in the 21st Century”. – 2017. – Vol. 19, No. 10. – P. 109–110. – EDN ZATOKV.
3. Dolgy, I. D. The relevance of implementing automatic blocking logical control of operational personnel actions in automated train traffic management systems / I. D. Dolgy, I. O. Alymov, A. O. Gorobets // Digital infocommunication technologies : collection of scientific papers, Rostov-on-Don, October 27, 2023. – Rostov-on-Don : Rostov State Transport University, 2023. – P. 380–384. – EDN AJAKND.
4. Komlichenko, V. N. Comparative analysis of various neural network architectures for regression problems / V. N. Komlichenko, V. A. Fedosenko, A. S. Kupreichik // Economics and Quality of Communication Systems. – 2025. – No. 1 (35). – P. 110–121. – EDN PAGCVM.
5. Pustinny, Ya. N. Solving the problem of a vanishing gradient using neural networks of long short-term memory / Ya. N. Pustinny // Innovations and Investments. – 2020. – No. 2. – P. 130–132. – EDN MRQIHM.
6. Lunt, A. An empirically-sourced heuristic for predetermining the size of the hidden layer of a multi-layer perceptron for large datasets. Lecture Notes in Computer Science / A. Lunt, S. Xu // LNAI. – 2016. – Vol. 9992. – P. 542–547. – DOI 10.1007/978-3-319-50127-7_47.
7. Cybenko, G. Approximation by superpositions of a sigmoidal function / G. Cybenko // Mathematics of Control, Signals, and Systems. – 1989. – Vol. 2, No. 4. – P. 303–314. – DOI 10.1007/bf02551274.
8. Goodfellow, Ya, Deep learning / Ya. Goodfellow, I. Bendjio, A. Courville. Moscow : DMK Press, 2018. 652 p. – ISBN 978-5-97060-618-6.
9. Data Science Stack Exchange. (2016). Number of parameters in an LSTM model. – URL: https://datascience.stackexchange.com/questions/10615/number-of-parameters-in-an-lstmmodel (date of access: 20.02.2026).
10. Hochreiter, S. Long short-term memory / S. Hochreiter, J. Schmidhuber // Neural Computation. – 1997. – Vol. 9 (8). – P. 1735–1780. – DOI 10.1162/neco.1097.9.8.1735
Review
For citations:
Alymov I.O. Analysis of parameters of basic neural network architectures in transport control systems. Vestnik Rostovskogo Gosudarstvennogo Universiteta Putej Soobshcheniya. 2026;(1):20-27. (In Russ.) https://doi.org/10.46973/0201-727X_2026_1_20
Views:
28
JATS XML
Email this article 