الفهرس 
INTRODUCTIONOnly 14 pages are availabe for public view
Abstract
This study presents a new model to predict the run life of Electrical Submersible pumps (ESP) installed in the Egyptian Western Desert oil fields. The study considers 363 ESP systems which were installed in 194 wells with 25 data points for each system. The study shows the methodology to evaluate the collected data and choose the most informative data to build the model. The developed model uses only 9 data points for each system based on statistical tests. Each data type is supported by previous literature showing its effect on ESP run life. The study relies on four different Machine Learning algorithms used in regression tasks. The results of each algorithm are evaluated with three accuracy matrices. The simple algorithm of Lasso Linear Regression has poor accuracy. The study shows that the model should rely on more complex algorithms due to ESP system complexity. The model with the highest accuracy was developed using the Neural Network algorithm. This model could predict the run life of the test data with Mean Absolute Error of ±34 days. It has an R2 value of 0.97. The model shows the importance of each data type in prediction results. The study shows that the model could evaluate ESP designs and optimize the running parameters. The model could also be used in planning workover schedules. The study shows the massive capability of Machine Learning as a cost-effective solution to oil field problems. The study results are promising in failure prediction and optimization of ESP operations.