SHORT-TERM STOCHASTIC FORECASTING OF THE DYNAMICS OF GAS FLOWS AND RESERVES IN THE UNIFIED GAS SUPPLY SYSTEM OF RUSSIA BASED ON RECURRENT NEURAL NETWORKS
UDC: 681.5:622.279
DOI: 10.33285/0132-2222-2021-9(578)-27-35
Authors:
KISLENKO NIKOLAY ANATOLIEVICH 1,2,
BELINSKY ALEXANDER VYACHESLAVOVICH2,
KAZAK ALEXANDER SOLOMONOVICH 2,
BELINSKAYA OLGA IGOREVNA
1 PJSC "Gazprom", St. Petersburg, Russian Federation
2 LLC "NIIgazeconomika", Moscow, Russian Federation
Keywords: gas transportation system, optimal control model, operational forecasting, stochastic forecasting, machine deep learning technologies, recurrent neural networks
Annotation:
The paper considers the urgent problem of operational forecasting of the dynamics of a large engineering system - the Unified Gas Supply System of Russia (UGSS). As the space of UGSS states, a continuous set of balance-flow variables was chosen that characterize the volume of gas transportation through gas transmission systems - UGSS fragments, as well as the volume of gas accumulated in the pipelines of these systems. The task is to model the dynamics of the UGSS functioning according to the given dynamics of the volumes of gas inflow and distribution in the UGSS. It is noted that adequate models should take into account the stochastic nature of the UGSS functioning, as well as nonlinear relationships between the parameters characterizing the state of the system. It is shown that the existing industry models do not provide an acceptable solution to this problem. A model based on machine deep learning technologies and modern architectures of recurrent neural networks is proposed. The training of the model, which has an applied value, was carried out on the actual data of daily gas balances for gas transmission enterprises in 2010-2020. The advantages of the model and its features that require further research are discussed. It is assumed that the proposed approaches will be developed in the form of new models of optimal control of the UGSS, which are also discussed in the work.
Bibliography:
1. Айвазян С.А., Фантаццини Д. Эконометрика-2: продвинутый курс с приложениями в финансах: учеб. - М.: Магистр; Инфра-М. - 2015. - 944 с.
2. Белинский А.В. Влияние газоснабжения и газификации на экономический рост российских регионов (эконометрический подход) // Газовая пром-сть. - 2018. - № S2(770). - С. 6-13.
3. Белинский А.В. Deep Reinforcement Learning в управлении крупными инженерными системами // Альманах "Искусственный интеллект". - М., 2020.
4. Кисленко Н.А., Казак А.С., Косова К.О. Регрессионный подход к прогнозу запаса газа в ГТС // Автоматизация, телемеханизация и связь в нефтяной промышленности. - 2020. - № 3(560). - С. 15-21. - DOI: 10.33285/0132-2222-2020-3(560)-15-21
5. Сарданашвили С.А. Расчетные методы и алгоритмы (трубопроводный транспорт газа). - М.: Нефть и газ, 2005. - 577 с.
6. Сухарев М.Г., Самойлов Р.В. Анализ и управление стационарными и нестационарными режимами транспорта газа: моногр. - М.: Изд. центр РГУ нефти и газа (НИУ) им. И.М. Губкина, 2016. - 399 с.
7. Belinsky A.V. Gas pipeline infrastructure and economic growth of Russian regions: panel cointegration analysis // E3S Web of Conferences. - 2019. - Vol. 102. Mathematical Models and Methods of the analysis and Optimal Synthesis of the Developing Pipeline and Hydraulic Systems, Irkutsk, Russia, June 16-22. - DOI: 10.1051/e3sconf/201910202002
8. Belinsky A., Afanasev V. Optimal Control of Energy Pipeline Systems Based on Deep Reinforcement Learning // "Smart Technologies" for Society, State and Economy (Lecture Notes in Networks and Systems, vol. 155). - Cham: Springer, 2021. - P. 1348-1355. - DOI: 10.1007/978-3-030-59126-7_148
9. Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation / Kyunghyun Cho, B. van Merrienboer, C. Gulcehre [et al]. - 2014. - URL: https://arxiv.org/pdf/1406.1078.pdf
10. Hinton G.E., Roweis S.T. Stochastic Neighbor Embedding // Advances in Neural Information Processing Systems. - Cambridge, MA, USA: The MIT Press, 2003. - Vol. 15. - P. 833-840.
11. Hochreiter S., Schmidhuber J. Long Short-Term Memory // Neural computation. - 1997. - Vol. 9, Issue 8. - P. 1735-1780. - DOI: 10.1162/neco.1997.9.8.1735
12. Jungsik Hwang. Modeling Financial Time Series using LSTM with Trainable Initial Hidden States. - 2020. - URL: https://arxiv.org/ftp/arxiv/papers/2007/2007.06848.pdf
13. Application of Long Short-Term Memory (LSTM) Neural Network for Flood Forecasting / Xuan-Hien Le, Hung Viet Ho, Giha Lee, Sungho Jung // Water. - 2019. - Vol. 11, Issue 7. - DOI: 10.3390/w11071387
14. Manibardo E.L., Laña I., Del Ser J. Deep Learning for Road Traffic Forecasting: Does it Make a Difference? - 2020. - URL: https://arxiv.org/pdf/2012.02260.pdf
15. Siami-Namini S., Tavakoli N., Siami Namin A. A Comparison of ARIMA and LSTM in Forecasting Time Series // 2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA), Orlando, FL, USA, Dec. 17-20. - 2018. - P. 1394-1401. - DOI: 10.1109/ICMLA.2018.00227
16. Sutskever I., Vinyals O., Le Q.V. Sequence to sequence learning with neural networks // Advances in neural information processing systems. - 2014. - P. 3104-3112. - URL: https://papers.nips.cc/paper/5346-sequence-to-sequence-learning-with-neural-networks.pdf
Back to issue