The presented article is devoted to the creation and testing of an ensemble probabilistic computational tool for operational forecasting of well flow in the short term. The ensemble includes models based on such physical and mathematical devices as: the equation of non-stationary filtration, material balance, Darcy's law and machine learning models. After making calculations by each model, their forecasts are combined into a single ensemble forecast. Each model makes a forecast based on historical information about the production and injection wells. The approach is based on the Monte Carlo method on Markov chains in the form of a separate probabilistic model using the Bayes formula. At the same time, the statistical weights of each model (the degree of confidence in each model) are determined in the form of a probability distribution based on the reliability of the retrospective component of the forecasts. The test results presented in this article were obtained on the basis of real field data of the deposit. Despite the shortcomings in the ensemble approach, the analysis of the tool use on real data showed that the proposed approach has a smaller average error on the forecast and a much smaller variance than each ensemble model separately. Forecasts were made for a short period of 30 to 90 days. Discretization of calculations in time was 1 day. The average value modulo the relative error for individual wells for the ensemble was 2.8% for liquid and 5.1% for oil, while the classical method of forecasting by the rate of decline gave an error of 24.5% and 24.3%, respectively.
1. Lake L.W., Petroleum engineering handbook. Volume I, Society of Petroleum Engineers, 2014.
2. Friedman J. et al., Regularization paths for generalized linear models via coordinate descent, Journal of Statistical Software, 2010, V. 33, DOI:10.1163/ej.9789004178922.i-328.7
3. Kim Seung-Jean et al., An interior-point method for large-scale L1-regularized least squares, IEEE Journal of selected topics in signal processing, 2007, V. 1, no. 4, DOI: 10.1109/JSTSP.2007.910971
4. Chen T., Guestrin C., XGBoost: A scalable tree boosting system, Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016, pp. 785–794, DOI:10.1145/2939672.2939785
5. Gabitova S.I., Davletbakova L.A., Klimov V.Yu. et al., A new method of decline curve forecasting for project wells on the base of machine learning algorithms (In Russ.), PRONEFT''. Professional'no o nefti, 2020, no. 4, pp. 69-74, DOI: 10.7868/S2587739920040102
6. Ranganathan A., The Levenberg-Marquardt algorithm, 2004.
7. 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-MS, 2006, DOI: 10.2118/95322-PA
8. Holanda R.W. et al., A state-of-the-art literature review on capacitance resistance models for reservoir characterization and performance forecasting, Energies, 2018, V. 11, no. 12, DOI: 10.3390/en11123368
9. Osvaldo M., Bayesian analysis with Python, Packt Publishing, 2018, 282 p.