Today one of the main competitive advantages in oil and gas industry is high-quality processing and analysis of large amounts of data for the subsequent tasks of production scheduling, aggregator loading, nominal production and well repair works. The use of machine learning methods is a relevant and a promising direction. However, one of the challenges is the impossibility of solving engineering problems using only machine learning algorithms or only physical and mathematical models. Using only one of the approaches is either more labor-intensive or allows for the possibility of non-physical solutions and high error values.
The paper presents new approaches for analyzing high-frequency data on the example of solving various problems of petroleum engineering. The proposed hybrid algorithms for data analysis, based on the use of statistical data processing and machine learning methods in conjunction with traditional hydraulic calculations, make it possible to significantly increase the value of incoming information by identifying and responding to problems in a timely manner, thereby improving field development efficiency without conducting additional studies. Algorithms allow indirect data, without direct flow measurements, to identify deviations from the planned mode of operation and errors in the operations of the metering infrastructure. The algorithms have been tested at Urals-Volga region. Effects have been obtained by the prevention of technological problems at wells, optimization of periodic short-term activation of ESP mode of artificial lifted wells, and automatic detection of problems in metering stations.
References
1. Volkov N.A., Dakhova E.Yu., Verification of field data and forecast model based on variational autoencoder in the application to the mechanized fund,
SPE-201936-MS, 2020, DOI: https://doi.org/10.2118/201936-MS.
2. Yudin E.V., Khabibullin R.A., Galyautdinov I.M. et al., Modeling of a gas-lift well operation with an automated gas-lift gas supply control system (In Russ.),
SPE-196816-MS, 2019, DOI: https://doi.org/10.2118/196816-MS.