ELECTRIC CENTRIFUGAL PUMP UNIT PREDICTION FAILURE BASED ON INDIRECT INDICATORS

This article presents existing solutions in the field of automation of the process of forecasting the electric submersible pump (ESP) unit failure using machine learning, and describes the advantages and disadvantages of existing solutions. A new approach to solving the problem of failure of pumps for oil production based on the anomaly detection is proposed. The last 15 days before ESP unit failure are taken as anomalies in operation. Using telemetry data to monitor and analyze the operation of ESP units opens up new opportunities in the field of optimization of oil production processes. Machine learning allows you to identify anomalies and predict possible equipment failures, which will significantly reduce the risks of critical situations at the field. The results of the research confirm that the analysis of indirect indicators of dynamic data obtained from sensors is a feasible tool for improving the reliability and efficiency of downhole equipment. Further development of this approach can lead to improvements in well operation processes and reduction in maintenance costs. The hypothesis that the well configuration affects the forecast is confirmed: the more signs for diagnosing the state of the system and predicting a breakdown, the higher the accuracy of the result. To expand the functionality, it is possible to implement an assessment of the potential for regulating the feed of the ESP, which would allow this prototype to both identify anomalies in operation and their cause, and recommend a well operating mode.