Performance Comparison of Isolation Forest and COPOD Algorithms for Anomaly Detection in Electrical Submersible Pumps

Asmara, Hygiano Paksi Widya
Maharsi, Dara Ayuda
Rachmanto, Rian

Journal of Geoscience, Engineering, Environment, and Technology
Special Issue from The 2nd International Conference on Upstream Energy Technology and Digitalization

The Electrical Submersible Pump (ESP) is a crucial technology in enhancing oil production, yet its performance can be compromised by anomalies that lead to operational disruptions and financial losses. Early detection of these anomalies is vital for minimizing risks and optimizing ESP lifespan. This study compares the performance of two machine learning algorithms—Isolation Forest and Copula-Based Outlier Detection (COPOD)—in identifying anomalies in ESP operational data. The study uses both long-term historical data and short-term period data from a well in Field X, focusing on key operational parameters such as amperes, frequency, voltage, discharge pressure, motor temperature, vibration, and gross rate. The results indicate that Isolation Forest outperforms COPOD in detecting anomalies, particularly in the presence of missing data. Short-term data detection yields clearer correlations between anomalies in different features, highlighting its advantage over long-term historical data. The findings underscore the importance of utilizing short-term operational data and demonstrate how anomaly detection algorithms can enhance ESP monitoring for improved performance and cost-efficiency.