Perpustakaan Digital ITB

Electrical Submersible Pumps (ESPs) are one of the most common artificial lift methods used in both offshore and onshore oil fields. With real-time monitoring systems, ESP data is updated every minute, creating a large volume of information. This study examines how to utilize this data to automatically optimize ESP performance and predict pump health. The objective is to increase oil production by optimizing the ESP frequency. An iterative process is applied, where the frequency is increased step-by-step. At the same time, key pump health parameters—such as intake pressure, discharge pressure, motor temperature, intake temperature, vibration, and motor load—are monitored. If these values remain within a safe range (as defined by experts), the frequency continues to increase. If not, the process stops. To predict pump health, machine learning models are used. The vibration-y model using XGBoost achieved an R² of 0.79, while the vibration-x model reached an R² of 0.84. The intake pressure model, using an Extra-Trees Regressor, achieved an R² of 0.98, while the discharge pressure model reached an R² of 0.83. For motor temperature, the model achieved an R² of 0.99. The virtual flowrate model using AdaBoost Regressor resulted in an R² of 0.90. Virtual flow rate is only predicted for the most recent data, while the forecast values are calculated using the affinity law. For motor load and intake temperature, an analytical method will be used. The method was tested on two wells. For Well-19, increasing the frequency from 107 to 120 Hz raised production from 12,768 to 14,319 BFPD, with an oil gain of 10.85 BOPD. For Well-11, the frequency increased from 106 to 120 Hz, improving flow from 13,033 to 14,754 BFPD, and gaining 12.05 BOPD. This study presents a novel approach to utilizing machine learning and pump behavior to enhance oil production automatically. While the results are good enough, further improvements, such as panel split cross-validation and advanced time-series models (e.g., LSTM), could enhance future work.