Publications

A failed Electrical Submersible Pump (ESP) well is typically identified when there is suboptimal flow or no flow reaching the surface. The process of reviving ESP well production can take weeks leading to huge unwanted deferment. Due to criticality for sustainable oil production, ESP failure prediction was highly envisaged. Therefore, the objective of this study is to develop a Predictive Analytics model based on ML algorithms using field sensor data, real time physics-based model calculated data, and well failure history to predict ESP well failure to identify failed event in advance.

Machine learning was used to predict when the next failure event will occur in ESP wells by correlating failure data with historical ESP readings. Using historical data failure from 2021- 2024 in 19 wells, it was found that the top performing model Random Forest achieves 86% accuracy in 10-fold cross validation.

This machine-learning technique can significantly enhance ESP surveillance by automating predictive solutions, facilitating better planning, and prioritization of workover candidates. These methods help reduce the time it takes to detect failure before unplanned shutdown event and improve ESP reliability. Additionally, they ensure the effective use of rigs, increase the run life of ESPs, and lead to significant savings from reducing total unscheduled ESP deferments. The benefit from this machine-learning implementation can reduce loss potential oil up to 17,000 barrel per year or saving cost $2,100,000 per year include rig. Machine learning for ESP failure prediction enhances monitoring and reliability, leading to better planning and reduced downtime.