Abstract
The development of a technology forecasting ESP failures is one of the key tasks of the oil industry today. Simple statistical methods do not allow to predict the equipment's failure and its cause, which would allow to select and conduct the planned preventive measure in time.
This paper describes the methodology of ESP failure prediction developed based on field data. The main stage of the methodology is the creation of an ensemble model of machine training, the application of which allows for the taking into account of many factors and extensive experience in operating ESPs contained in field data, so considerable time has been devoted to working with them.
The uniqueness of the presented work is the use of field information on well operation in Western Siberia, accumulated for the last 5-7 years from the discrete measurements of one day, in which there was a significant amount of information omission. To improve the quality of forecasting, it required multilevel pre-processing of initial data and formation of additional analytical features, which allowed improvement of the quality of predictions.
The study of the developed model for the prediction of the ESP failures at different prediction horizons was carried out and it was determined that the quality of prediction increases when the interval of prediction decreases. The developed model shows the best results when predicting the ESP failure 1-3 days before its actual stop.
The groups of parameters, on which the duration of failure-free operation of pumping equipment in a well depends to a great extent, have been determined this is information about the well operation mode, information about ESP operation parameters, dynamic information about complicating factors of the ESP operation and additional analytical parameters describing the processes occurring in the system "Wellbore – ESP – Tubing".
The ESP failure prediction model, the development of which is described in this work, can become the basis of the decision support system, its implementation in production will allow to identify in advance the cases of possible downhole pumping equipment failure and to take timely measures based on this information.
Based on the obtained results and high relevance of the problem under consideration, the main directions for further improvement of the developed methodology have been defined.
