Utilization of Predictive Data Analytics in Enhanced Oil Recovery Optimization and Decision Making

Abstract

Enhanced oil recovery process has various optimization facets; each has its own vital role to play in improving the overall efficiency of such a critical process in the reservoir development plan. The first and foremost step in any enhanced oil recovery project is the selection of the proper technique, from various enhanced oil recovery techniques, that might be more efficient in the specific candidate reservoir with its unique rock and fluid properties. Substantial amount of reservoir data has been accumulated over decades from many fields worldwide. Oil and gas companies demonstrating a growing interest in leveraging these data, taking advantage from the fast advancements in computer science and artificial intelligence approaches. Utilization of machine learning and other data analytics techniques has emerged as a promising approach to extract valuable insights from these data, consequently enhancing decision-making, optimizing various processes, and improving outcomes. In this research, we focused on classifying enhanced oil recovery methods according to fluid properties and reservoir characteristics by employing a predictive data analytics approach, namely machine learning model, to optimize the EOR screening for a candidate reservoir. Based on the model evaluation results, ensemble algorithms, notably random forest (RF), outperformed other algorithms, demonstrating relatively higher classification accuracy and evaluation scores. Hence, the research delved deeper into the selected model, the random forest algorithm, and discussed model uncertainty and optimization techniques. Although the employed methodology in this research is valid and provides a robust framework for utilizing predictive data analytics in EOR screening, it is crucial to acknowledge that this approach has certain pitfalls that might originate from data quality, inherent complexities in oil reservoir systems, or the selected capabilities of the algorithm. Therefore, further optimization is highly recommended before the model can be reliably adopted and implemented in the practical applications. Future works could focus on improving data quality and input parameters or even, utilizing of more advanced data analytics and machine learning algorithms that can better capture the nuances of enhanced oil recovery screening problem.