Enhancing electrical submersible pump (ESP) reliability through integrated data-driven machine learning framework: A case study from Egypt

• In this research paper: • Comprehensive review of ESP failure prediction methods and challenges. • Detailed analysis of ESP behavior enables proactive failure prevention. • Static features used for effective long-term ESP runlife prediction. • Classification model identifies ESP failure causes with 96% precision. • Time-independent features provide stable input for long-term forecasting. Electrical Submersible Pump (ESP) is the most competent and consistent method for medium-to-high production rates. However, with the rising number of ESP wells, engineers strive to improve their performance and attain a longer runlife. While most studies rely on time-series data of downhole sensors for real-time ESP monitoring, this can be challenging for proactive maintenance. This study introduces a different approach using static, time-independent features from reservoir characteristics, production data, wellbore trajectory, and equipment specifications. Static features, which remain constant over time, offer valuable insights for long-term forecasting; hence they will be more appropriate to depend on to predict the failure long before it happens. By analyzing these features, ESP failures have been successfully predicted in onshore fields of Egypt with an acceptable level of confidence, enabling more efficient operations. The solution used an integrated database, in addition, the system incorporated a machine learning regression model to predict ESP lifespan and a classification model to identify the most likely causes of pulls. Machine learning models had been trained on ESP activities of 231 wells with a dataset of 676 system installations over 14 years. The regression model exhibited strong predictive accuracy of pump expected runlife, with a mean absolute error of 17 days. The classification model achieved a high precision of 96%. The results offer valuable guidance for workover planning, performance optimization, and inventory management, especially considering current supply chain challenges. The system architecture and workflow pave the way for further development and even more precise diagnostics and prediction.