inttransИнтеллектуальный транспортIntelligent transport3033-6007АО «НИИАС»inttrans-26Research ArticleИСКУССТВЕННЫЙ ИНТЕЛЛЕКТ И МАШИННОЕ ОБУЧЕНИЕПрименение глубокого обучения для предсказания отказов подвижного составаApplication of deep learning for rolling stock failure predictionКомисарчукА. В.KomisarchukA. V.

Комисарчук Алексей Васильевич, главный системный аналитик

Москва

Aleksey V. Komisarchuk, system architect, M.Sc.

Moscow

komisarchuk.aleksei@vniizht.ruЕлисеевИ. А.EliseevI. А.

Елисеев Игорь Александрович, начальник отдела, к.т.н.

Москва

Igor А . Eliseev, department chief, cand. sci. (eng.)

Moscow

eliseev.igor@vniizht.ruСидоровА. В.SidorovA. V.

Сидоров Алексей Викторович, руководитель проектов

Москва

Aleksey V. Sidorov, project manager

Moscow

Sidorov.Alexey@vniizht.ruАО «ВНИИЖТ»РоссияRailway Research InstituteRussian Federation20251109202604(36)1424Copyright © Комисарчук А.В., Елисеев И.А., Сидоров А.В., 20262026Комисарчук А.В., Елисеев И.А., Сидоров А.В.Komisarchuk A.V., Eliseev I.А., Sidorov A.V.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://www.intelligent-transport.ru/jour/article/view/26

Эта работа открывает серию исследований, объясняющих необходимость и возможность прогрессивной диагностики технического состояния железнодорожного подвижного состава, и предлагает принципиально новое решение этой проблемы на основе современного интеллектуального алгоритма глубокого обучения. Здесь сформирована абстрактная концепция аномальности данных, дана формальная математическая постановка задач поиска и прогнозирования отказов промышленного оборудования, описаны основные компоненты программной реализации решения этих задач с применением модели рекуррентной нейросети (LSTMавтоэнкодера), а также приведены оценка качества работы и аргументация в пользу выбора глубокого обучения среди актуальных методов в области искусственного интеллекта.

This paper begins a series of publications, making a case and opportunity for creation of advanced diagnostics of technical condition of the railway rolling stock and provides essentially new artificial intelligence (AI) solution based on deep learning techniques. Here, the abstraction for anomalous data is formed, statements of industrial equipment failure detection and forecasting problems are set, core components of software implementation of the solution algorithm using the recurrent network (LSTM autoencoder) are described as well as the quality assessments and the reasoning for choosing deep learning over the rest AI methods.

искусственный интеллектнейросетьпрескриптивная аналитикааномалияотказжелезнодорожный подвижной составartificial intelligenceneural networklong short-term memoryprescriptive analyticsanomalyfailurerailway rolling stockReferencesD. Hawkins, Identification of Outliers. London: Chapman and Hall, 1980D. Hawkins, Identification of Outliers. London: Chapman and Hall, 1980J. J. Hopfield, “Neural networks and physical systems with emergent collective computational abilities,” Proc. Nat. Acad. Sci. U.S.A., vol. 79, no. 8, pp. 2554–2558, 1982.J. J. Hopfield, “Neural networks and physical systems with emergent collective computational abilities,” Proc. Nat. Acad. Sci. U.S.A., vol. 79, no. 8, pp. 2554–2558, 1982.S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Comput., vol. 9, no. 8, pp. 1735–1780, 1997.S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Comput., vol. 9, no. 8, pp. 1735–1780, 1997.N. Srivastava, E. Mansimov, and R. Salakhutdinov, “Unsupervised Learning of Video Representations using LSTMs,” in Proc. Int. Conf. Mach. Learn. (ICML), 2016, pp. 843–852.N. Srivastava, E. Mansimov, and R. Salakhutdinov, “Unsupervised Learning of Video Representations using LSTMs,” in Proc. Int. Conf. Mach. Learn. (ICML), 2016, pp. 843–852.J. F. Kolen and S. C. Kremer, A Field Guide to Dynamical Recurrent Networks. New York: Wiley-IEEE Press, 2001.J. F. Kolen and S. C. Kremer, A Field Guide to Dynamical Recurrent Networks. New York: Wiley-IEEE Press, 2001.I. Sutskever, O. Vinyals, and Q. V. Le, “Sequence to Sequence Learning with Neural Networks,” in Adv. Neural Inf. Process. Syst. (NIPS), 2014, pp. 3104–3112.I. Sutskever, O. Vinyals, and Q. V. Le, “Sequence to Sequence Learning with Neural Networks,” in Adv. Neural Inf. Process. Syst. (NIPS), 2014, pp. 3104–3112.A. Sagheer and M. Kotb, “Unsupervised pre-training of a deep LSTM-based stacked autoencoder for multivariate time series forecasting problems,” Sci. Rep., vol. 9, p. 19038, 2019.A. Sagheer and M. Kotb, “Unsupervised pre-training of a deep LSTM-based stacked autoencoder for multivariate time series forecasting problems,” Sci. Rep., vol. 9, p. 19038, 2019.N. Whiteley, A. Gray, and P. Rubin-Delanchy, “Statistical exploration of the Manifold Hypothesis,” J. Amer. Stat. Assoc., 2025. (Early Access).N. Whiteley, A. Gray, and P. Rubin-Delanchy, “Statistical exploration of the Manifold Hypothesis,” J. Amer. Stat. Assoc., 2025. (Early Access).A. Vaswani et al., “Attention is all you need,” in Adv. Neural Inf. Process. Syst. (NIPS), 2017, pp. 5998–6008.A. Vaswani et al., “Attention is all you need,” in Adv. Neural Inf. Process. Syst. (NIPS), 2017, pp. 5998–6008.K. Besbes et al., “Sizing of an electric powertrain using a hybrid model combining Simcenter Amesim and Modelica components,” in Proc. IEEE Veh. Power Propuls. Conf. (VPPC), 2020, pp. 1–6.K. Besbes et al., “Sizing of an electric powertrain using a hybrid model combining Simcenter Amesim and Modelica components,” in Proc. IEEE Veh. Power Propuls. Conf. (VPPC), 2020, pp. 1–6.

The authors declare that there are no conflicts of interest present.