English abstract
The life cycle of an oil production well can be divided into several stages, from exploration
to abandonment. Each stage serves a specific purpose and involves different activities and
considerations. Production cycle is one of the important stage where the well begins to
flow and produce economically. At this stage, production rates are monitored, and
reservoir management strategies are implemented to maximize the recovery and minimize
well decline. At the initial production, most of the wells are naturally flowing by rely on
natural reservoir pressure to bring oil to the surface. Most of these wells produced at lower
operating costs. By the time, the reservoir pressure is declined and considered as
insufficient to bring the oil to the surface effectively. Here, the Artificial Lift systems are
implemented to enhance production rates and deplete the natural pressure. They help
maintain or increase production levels by providing the necessary energy to bring oil to
the surface. These wells tend to have higher operational costs due to the energy required
to power the artificial lift mechanisms, and they may require more maintenance and
monitoring to ensure efficient and consistent production. Selecting the appropriate
artificial lift system is an important decision in this process, as it significantly impacts
well productivity and operational costs. Traditional methods for choosing the optimal lift
system often depends on empirical data and experts knowledge, making the process timeconsuming and sometimes suboptimal which can lead into higher operational costs. To
address this challenge, the present study proposes a development and implementation of
Machine Learning (ML) model to predict the most suitable artificial lift system for
individual oil wells in South Oman Fields. The study began by collecting the wells
completion and production data for almost 995 horizontal produces operated with four
types of Artificial Lift Systems (ALS) in the field. The data has been cleaned to build up
the model and validated it to make it ready for the deployment. Feature engineering and
data preprocessing are conducted to ensure the quality and relevance of input data. By
applying six machine learning classification algorithms, including Decision Tree, KNearest Neighbors, Random Forest, Support Victor Machines, Naïve Bayes and Logistic
Regression we learned to identify patterns and relationships between well attributes and
the optimum lift systems. The Thesis aims to improve the decision-making process by
providing a data-driven approach that can assist operators and fresh engineers in selecting
the optimum artificial lift system for a given well. The results of this research were
excellent as four algorithms scores an accuracy higher than 95% at a split ratio of (70:30,
Train: Test) which makes the model reliable be used in future applications.
