ИСТИНА |
Войти в систему Регистрация |
|
Интеллектуальная Система Тематического Исследования НАукометрических данных |
||
Objectives/Scope: The proposed paper will describe the results and operation principles of the developed AI-based solution for the optimization of operating modes of electrical submersible pumps (ESP). The solution allows exploration and production companies to increase the oil flow rate from ESPs without additional capital investments. The solution is based on the technical possibility of ESP mode adjustments in a certain range of motor frequencies and control types. Methods, Procedures, Process: The hypothesis is that there is a certain ESP operation mode in each period of time which ensures the maximum daily oil and gas flow rate in the long term. The proposed solution uses statistical methods to ensure data sufficiency and stability and employs custom algorithms to conduct problem-specific data preprocessing. The "well-reservoir" model is the combination of physical (i.e., pump model) and machine learning (ML) models. The ML part of the solution was designed for the short- and long-term predictions of well-reservoir system response. However, the model faced constraints such as current frequency, load and amperage. Results, Observations, Conclusions: The following results were delivered: data collection and preprocessing pipeline was built, a hybrid physical-ML prediction model was designed and trained, and a numeric optimization model which uses outputs of the hybrid prediction model was designed to recommend multi-well operation modes. The solution has been successfully deployed and tested for 6 months on 500 oil wells in Western Siberia. The designed ESP optimal control solution could enhance the efficiency of oil extraction by boosting oil well production rates by 1.5% with no additional capital investment, as proven during the production phase. Thus, the proposed "reservoir-well-pump" system modeling approach showed efficiency and superiority over the previous control methods used by oilfield operators. Novel/Additive Information: To the authors' best knowledge, the present paper is the first work covering research, development and production testing of hybrid physical-ML methods in the optimal control of ESPs with respect to optimization of oil flow. Application of some known statistical methods to the problem of provision of reliable data to the model is additive to the current body of knowledge.