One of the crucial components of any oil company successful development is high-quality processing and analysis of a large amount of available data. It is necessary for subsequent solution of forecasting and planning problems for oil, liquid and associated gas production. Among all of the actively developing digital technologies, one should specifically note the machine learning methods that have a great potential for solving time series forecasting problems on a whole variety of datasets of different nature. However, practice has shown that most of the oil engineering challenges cannot be solved effectively using only machine learning algorithms or only mathematical models of physical processes. Solving these problems using only one of these approaches is either more difficult / time-consuming and requires a lot of additional data and deep understanding of physics of the process (in the case of description of all the processes in the system by a complete mathematical model), or allows for a possibility of non-physical solutions and high error (in the case of using only machine learning approaches). In order to resolve these issues, the hybrid method of simplified physical model combined with machine learning model is presented in this paper. The proposed hybrid approach combines machine learning methods and the basic simplified pseudo-2D physico-mathematical model, and allows to minimize calculation errors arising from the impossibility of higher detalization of the basic model using implicit dependencies obtained by the machine learning model, which adjusts the main forecast. Also, the proposed hybrid approach allows, if necessary, to introduce new control parameters that are not taken into account by the physico-mathematical model, but can have a significant influence on the final result. The paper shows that the quality of adaptation to actual data and the quality of the production forecast satisfy the requirements for full-scale hydrodynamic 3D models.
References
1. Yakovlev V.V., Khasanov M.M., Sitnikov A.N., et al., The direction of cognitive technologies development in the Upstream Division of Gazprom Neft Company (In Russ.), Neftyanoe Khozyaystvo = Oil Industry, 2017, no. 12, pp. 6-9, DOI: 10.24887/0028-2448-2017-12-6-9.
2. Zotkin O., Osokina A., Simonov M. et al., A novel approach to refinment reservoir proxy model using machine-learning techniques (In Russ.), SPE-198411-MS, 2019, DOI: 10.2118/198411-MS
3. Teplyakov N., Slabetskiy A., Sarapulov N. et al., Application of machine learning methods for modeling the current indicators of operating wells stock of PJSC Gazprom Neft (In Russ.), SPE-191585-18RPTC-MS, 2018.
4. Khatmullin I.F., Tsanda A.P., Andrianova A.M. et al., Semi-analytical models for calculating well interference: limitations and applications (In Russ.), Neftyanoe khozyaystvo = Oil Industry, 2018, no. 12, pp. 38–41, DOI: 10.24887/0028-2448-2018-12-38-41.
5. Hochreiter S., Schmidhuber J., Long short-term memory, Neural computation, 1997, no. 9, pp. 1735-1780, DOI: 10.1162/neco.1997.9.8.1735.
6. Yousef A.A., Gentil P.H., Jensen J.L., Lake L.W., A capacitance model to infer interwell connectivity from production and injection rate fluctuations, SPE-95322-PA, 2009, DOI: 10.2118/95322-PA.
7. Sayarpour M., Zuluaga E., Kabir C.S. et al., The use of capacitance–resistance models for rapid estimation of waterflood performance and optimization, SPE-110081-MS, 2007.
8. Buzinov I.U., Umrikhin S.N., Issledovanie neftyanykh i gazovykh skvazhin i plastov (Investigation of oil and gas wells and reservoirs), Moscow: Nedra Publ., 1984, 269 p.
9. Stock J.H., Watson M.W., Heteroskedasticity‐robust standard errors for fixed effects panel data regression, Econometrica, 2008, V. 76(1), pp. 155–174, https://doi.org/10.1111/j.0012-9682.2008.00821.x
