«Automation and Informatization of the fuel and energy complex»

Neural-hydraulic networks for assessing the production capacity of the existing and developing gas transmission systems

UDC: 519.673
DOI: -

Authors:

¹ Research Institute of Gas Economy, Moscow, Russia

Keywords: modeling, capacity, gas transmission system, graph neural network, neural hydraulic network

Annotation:

The development of rational plans for the operation and development of gas supply systems is based on modeling gas flows in the Unified Gas Transmission System (GTS) of Russia. The planned flow patterns should be technologically feasible, i.e. the GTS should have sufficient reserves of production capacity. In case of "bottlenecks" presence in the GTS, technical and organizational solutions are developed to eliminate them. The assessment of the technically feasible throughput capacity (TFC) of existing GTS sections as well as the search for rational solutions of reconstruction and capacity development is traditionally performed by experts based on the results of multivariate hydraulic and technical-economic calculations. The problem is combinatorial. For example, long-term planning requires making an informed choice between numerous capacity expansion options, while repair planning requires selecting the scope of work and its sequence. These problems solution is time-consuming because the calculation of technically feasible throughput capacity (TFC) of the existing GTS sections takes considerable time. The authors of the article present a new method for estimating the technically feasible throughput capacity (TFC) of the existing and developing GTS sections. The method is based on the use of a neural-hydraulic network – a specialized neural network that approximates the set of solutions to this problem. The proposed network is based on the architecture of a physically-informed graph neural network and allows modeling the thermal hydraulic operational modes of GTS sections. Using an incidence matrix, the characteristics of GTS section objects and boundary conditions as input, the neural-hydraulic network predicts the TFC of this section. The application of differentiable physics concepts allowed for the creation of a fully auto-differentiable implementation of the model, taking into account various nuances of the physical relationships between the modeled parameters, known from hydraulic circuit theory. A novel neural network training method is proposed, based on the concepts of global gradient and extended Lagrangian methods. The studies of its generalization ability have shown that the trained neural network provides acceptable (for evaluation and pre-design calculations) TPS assessment results for a wide range of real-world sections of the hydraulic structure. Furthermore, it exhibits high computational performance, exceeding traditional TPS assessment algorithms by 3–4 orders of magnitude. It is noted that the primary purpose of neural hydraulic networks is not to replace hydraulic models, but to provide an effective instrument for exploring a wide range of alternative options for the operation and development of gas supply systems.

Bibliography:

1. Merenkov A.P., Khasilev V.Ya. Teoriya gidravlicheskikh tsepey. – M.: Nauka, 1985. – 279 s.
2. Sukharev M.G., Stavrovskiy E.R. Optimizatsiya sistem transporta gaza. – M.: Nedra, 1975. – 277 s.
3. Sukharev M.G., Samoylov R.V. Analiz i upravlenie statsionarnymi i nestatsionarnymi rezhimami transporta gaza. – M.: Izdatel'skiy tsentr RGU nefti i gaza (NIU) imeni I.M. Gubkina, 2017. – 399 s.
4. Sardanashvili S.A. Raschetnye metody i algoritmy (truboprovodnyy transport gaza). – M.: Izd-vo "Neft' i gaz" RGU nefti i gaza imeni I.M. Gubkina, 2005. – 577 s.
5. Sukharev M.G., Popov R.V. Sostoyanie i perspektivy sovershenstvovaniya matematicheskogo i komp'yuternogo obespecheniya v sfere operativnogo upravleniya i srednesrochnogo planirovaniya rezhimov krupnomasshtabnykh gazotransportnykh sistem // Nauchno-tekhnicheskiy sbornik Vesti gazovoy nauki. – 2018. – № 2(34). – S. 4–13.
6. Ob obosnovannosti primeneniya, sovremennom sostoyanii i nekotorykh perspektivakh razvitiya neyrosetevykh modeley Edinoy sistemy gazosnabzheniya Rossii / N.A. Kislenko, A.V. Belinskiy, A.S. Kazak, O.I. Belinskaya // Avtomatizatsiya i informatizatsiya TEK. – 2022. – № 5(586). – S. 6–17. – DOI: 10.33285/2782-604X-2022-5(586)-6-17
7. Razrabotka i aprobatsiya metodicheskikh podkhodov i tsifrovykh tekhnologiy neyrosetevogo proksi-modelirovaniya ustanovivshegosya dvukhfaznogo techeniya mnogokomponentnoy smesi v sistemakh sbora i promyslovoy podgotovki gaza (na primere Chayandinskogo NGKM) / A.V. Belinskiy, V.A. Marishkin, V.V. Samsonova, P.V. Pyatibratov // Avtomatizatsiya i informatizatsiya TEK. – 2024. – № 4(609). – S. 44–59.
8. O novom metode tsifrovogo modelirovaniya nestatsionarnykh rezhimov techeniya gaza v magistral'nykh gazoprovodakh s primeneniem neyronnykh operatorov / A.V. Belinskiy, D.V. Gorlov, I.A. Pyatyshev, A.E. Titov // Gazovaya promyshlennost'. – 2024. – № 5(865). – S. 54–66.
9. Belinskiy A.V. Imitatsionnaya nestatsionarnaya gibridnaya model' krupnoy gazotransportnoy sistemy s apparatnym uskoreniem vychisleniy na graficheskikh protsessorakh // Avtomatizatsiya i informatizatsiya TEK. – 2025. – № 2(619). – S. 34–47.
10. Ramsundar V., Krishnamurthy D., Viswanathan V. Differentiable Physics: A Position Piece. – 2021. – DOI: 10.48550/arXiv.2109.07573
11. Blondel M., Roulet V. The Elements of Differentiable Programming. – 2024. – DOI: 10.48550/arXiv.2403.14606
12. STO Gazprom 2-3.5-051-2006. Normy tekhnologicheskogo proektirovaniya magistral'nykh gazoprovodov. – M.: IRTs Gazprom, 2006. – VIII, 196 s.
13. Korel'shneyn L.B., Pashenkova E.S. Opyt ispol'zovaniya metodov global'nogo gradienta i dekompozitsii pri raschete ustanovivshegosya neizotermicheskogo techeniya zhidkostey i gazov v truboprovodakh // Truboprovodnye sistemy energetiki. Matematicheskoe modelirovanie i optimizatsiya. – Novosibirsk: Nauka, 2010. – S. 103–114.
14. Kingma D., Ba J. Adam: A Method for Stochastic Optimization // 3rd International Conference on Learning Representations (ICLR 2015), San Diego, CA, USA, May 7–9, 2015. – DOI: 10.48550/arXiv.1412.6980