Method for constructing an adaptive suboptimal stationary train traffic controller based on artificial neural networks

The article discusses the modern methodology for performing the synthesis of a suboptimal train controller for the purpose of energy saving. The existing methods of optimal traction control have a number of disadvantages, the main one of which is the lack of direct use in the control program of the data obtained during train operation. Mathematical models used to solve the op- timal problem can be used correctly only in the case of sufficient adequacy. Adequacy check is not part of the known methods of optimal control theory. To eliminate this drawback, it is proposed to use the method of optimal (suboptimal) traction calculations based on artificial neural networks. It improves the accuracy of traction calculations, which is especially important in the aspect of considering energy savings, while reducing the need for computing power. When using this method, it is possible not only to achieve results close to the classical Bellman method, but also to train or verify the network using the recorded data. The article discusses the process of creating and training an artificial neural network based on model data to solve the problem of suboptimal control. The train motion modes obtained by Bellman's method were used as reference data for training the neural network. The presented comparative results of the two methods show the applicability of artificial neural networks for solving applied problems of train traction with the possibility of continuous learning, including the use of trip data, which can be directly included in the training or testing set.

1. Baranov L. A., Erofeev E. V., Meleshin I. S., Chin' L. M. Optimizatsiya upravleniya dvizheniem poezdov. Ucheb. posobie [Optimization of train traffic control. Textbook]. Moscow, MIIT Publ., 2011, 164 p.

2. Erofeev E. V. Vybor optimal'nogo rezhima vedeniya poezda na ATsVM s primeneniem metoda dinamicheskogo programmirovaniya [The choice of the optimal mode of running a train on a digital computer using the method of dynamic programming]. Trudy MIIT [Proceedings of the MIIT], 1967, Vol. 228, pp. 16–30.

3. Bellman R., Dreyfus S. Prikladnye zadachi dinamicheskogo programmirovaniya [Applied problems of dynamic programming. English translation by N.M. Mitrofanova]. Moscow, Nauka Publ., 1965, 458 p.

4. Yurenko K. I. Raschet energooptimal'nykh rezhimov dvizheniya perspektivnogo podvizhnogo sostava metodom dinamicheskogo programmirovaniya [Calculation of energy-optimal modes of motion of perspective rolling stock by the method of dynamic programming]. Izvestiya vuzov. Elektromekhanika [University news. Electromechanics], 2013, no. 3, pp. 78–82.

5. Malakhov S. V., Kapustin M. Yu. Realizatsiya dinamicheskogo programmirovaniya s uchetom effektivnogo perescheta v bortovoy sisteme [Implementation of dynamic programming taking into account effective recalculation in the onboard system]. Science and Technology in Transport, 2020, no. 2, pp. 74–80.

6. Muginshteyn L. A., Ilyutovich A. E., Yabko I. A. Energooptimal'nye metody upravleniya dvizheniem poezdov. Sb. nauch. tr. OAO “VNIIZhT” [Energy-optimal methods of train traffic control. Proc. of the JSC “VNIIZhT”]. Moscow, Intext Publ., 2012, 80 p.

7. Yurenko K. I., Savos'kin A. N., Fandeev E. I. Issledovanie optimal'nykh traektoriy dvizheniya poezda s uchetom sluchaynykh faktorov [Investigation of optimal trajectories of the train motion taking into account random factors]. Trudy 3-y nauch.-tekhn. konf. “Intellektual'nye sistemy upravleniya na zheleznodorozhnom transporte ISUZhT-2014” (Moskva, 18 noyabrya 2014 g.) [Proc. of the 3rd scientific and technical conference “Intelligent control systems in railway transport ISUZHT-2014” (Moscow, November 18, 2014)]. Moscow, JSC “NIIAS” Publ., 2014, pp. 211–214.

8. Kolmogorov A. N. O predstavlenii nepreryvnykh funktsiy neskol'kikh peremennykh v vide superpozitsiy nepreryvnykh funktsiy odnogo peremennogo i slozheniya [On the representation of continuous functions of several variables in the form of superpositions of continuous functions of one variable and addition]. Doklady AN SSSR [USSR’ Science Academy reports], 1957, Vol. 114, no. 5, pp. 953–956.

9. Hecht-Nielsen R. Kolmogorov's Mapping Neural Network Existence Theorem. IEEE First Annual Int. Conf. on Neural Networks. San Diego, 1987, Vol. 3, pp. 11–13.

10. Khaykin S. Neyronnye seti: polnyy kurs [Neural networks: Complete course]. Izd. 2-e, ispr. [Edition 2nd, revised]. Moscow, St. Petersburg, Dialektika Publ., 2019, 1104 p.