Machine Learning methods to predict protential errors in electrical submersible pump operation

Abstract

The electrical submersible pump (ESP) has long been one of the most effective artificial-lift methods to improve the production rates and extend the life of oil wells. Although ESP has many great advantages, the monitoring and repairing process of ESP poses some problems. In order to minimize the economic losses caused by ESP errors, many methods have been proposed to predict the abnormalities of ESP system before making replacement and repair plans for ESP. The petroleum engineer must choose the most effective, time-saving and economical method. In recent years, AI-Artificial Intelligence has been extremely strongly developing, in which ML-Machine Learning is a prominent achievement. With ML, the prediction of ESP’s error and operating trend is no longer as difficult as before. ML uses the historical data set of previous ESPs to accurately and easily forecast possible future events. As the result, the cost and the time to repair and replace the ESP system may reduce. This research mentioned two most effective and popular ML methods: Extreme Gradient Boosting (XGboosting) and Artificial Neural Network (ANN). These two methods were used to predict errors that may arise during the ESP operation. The results of the two algorithms were compared together to find out the more optimal model in ESP error prediction. In addition, the research also evaluated the influence of each parameter on the error of ESP with the aim of preventing hidden ESP errors.