Application of Machine Learning for EOR Screening: A Worldwide Perspective

Abstract

The successful implementation of enhanced oil recovery (EOR) methods is highly dependent on reservoir properties, which determine the most suitable EOR techniques. While deploying EOR can be costly, it has the potential to improve the reservoir recovery factor significantly. However, the application of EOR requires significant upfront financial investment, has high running expenses, and is associated with operational risks. A reliable method for EOR screening is necessary in light of recent advances and new EOR techniques that have been developed along the years. The development of the classical look-up tables for EOR screening can be outdated missing to include the most recent advances. However, performing EOR screening is challenging, and conventional screening analysis is prone to expert-induced bias. It can be time-consuming, which in some cases can be detrimental to the whole objective of implementing EOR methods, incrementing the recovery on reserves. A comprehensive strategy for determining the most effective EOR technology based on reservoir conditions is currently lacking.

As an alternative to conventional screening, this project proposes a comprehensive approach for EOR screening using Artificial Intelligence (AI). In this project, a machine learning (ML) algorithm was applied to relate key reservoir parameters, such as temperature, depth, oil gravity, lithology, porosity, permeability, thickness, reservoir pressure, oil saturation, and salinity to the different EOR methods, including chemical, gas, and thermal EOR. A large worldwide database from the literature and recent EOR surveys was collected and fed into both supervised- and unsupervised-learning techniques to assess the best EOR strategy. Finally, after validation, selected algorithms were evaluated on reservoir data and applied to other oil fields not considered for the machine learning model training. All the ML models were implemented in Python using open-source machine learning libraries.

Date of Award	Dec 2022
Original language	American English
Supervisor	WALEED ALAMERI (Supervisor)