In the modern era, data plays a crucial role in technological advancements, including in the field of Petroleum
Engineering. Production data, as a critical dataset, can be utilized to identify the economic viability of an oil and
gas field. Production optimization must be conducted to ensure the sustainable operation of the field. One method
used in optimization and production enhancement in the oil and gas industry is the Artificial Lift system. A
commonly used artificial lift method to boost oil production, particularly in mature fields, is the Electrical
Submersible Pump (ESP). In mature fields, ESP failures can lead to a decline in oil production. Therefore,
performance analysis of ESPs is essential to maximize their lifespan and achieve optimal profitability. One
analysis method that can be employed is using Artificial Intelligence (AI) to help to predict the ESP's Remaining
Useful Life (RUL Pred).
This research applies a combination of data digitization methods and optimized machine learning models for
predicting the remaining useful life (RUL) of Electrical Submersible Pumps (ESP). Additionally, this study
introduces an Early Warning System based on forecasting the predicted RUL of ESP, which provides a preventive
advantage in managing ESP performance.
In this research, data was obtained through digitization of data from four ESP. The data collection process involved
digitization, data normalization using min-max normalization as a pre-processing step, followed by training,
testing, and data validation to evaluate and obtain a model capable of accurately predicting ESP RUL Pred.
Predictions from this model yield an Early Warning System regarding ESP conditions, providing operators with
time to take necessary preventive measures.
From performing several case studies, the most accurate machine learning model for predicting RUL Pred ESP is
the Gradient Boosting model, as shown by multiple case studies. This model consistently produced low error
values, with an average MAE of 0.019, MSE of 0.003, and RMSE of 0.041, alongside an R² value close to 1 in
several cases. This study also investigates the importance factors that influencing the prediction of RUL Pred ESP.
Feature importance analysis identified that the Motor Temperature (Tm), and Pump Intake Pressure (PIP) are the
most important features that affect the prediction, affecting up to above 0.2 on decreasing R2 value. Meanwhile
parameters like Intake Temperature (Ti), Month, and Ampere show not significantly high affect in predicting the
RUL Pred ESP as affecting below 0.15 in decreasing the R2 value. Consequently, it is recommended that the
analysis can be complemented with other methods to comprehensively assess parameter importance in predicting
RUL Pred ESP.
However, for the application of this machine learning model to the dataset outside of this research, further studies
are needed to evaluate the distribution, quantity, and variables of the data to ensure that the model can generalize
and produce accurate predictions.