Optimization of Oil Recovery Factor Using Machine Learning Methods Integrated with Eclipse Simulator

Abstract

As the world grapples with the challenges of sustainable energy, oil and gas remain the main sources of the world's energy supply. While, global oil consumption has increased significantly, the accessible reserves of oil and gas resources have decreased as a result of the usage of common light oils. To maximize extraction from known sources, secondary recovery procedures, often known as EOR, - enhanced oil recovery- are used when relying exclusively on reservoir pressure is no longer adequate. EOR projects depend on variety of factors including economic factors, crude oil market price, and the investment from individuals and firms willing to take on the risk. To optimize extraction beyond reservoir pressure and maximize reservoir recovery and profits, several methods for EOR have been proposed including water injection and polymer and surfactant injection. This paper examines the use of three distinct machine learning simulation models- the random forest regressor, the gradient boost regression, and the KNN Regression- to investigate these EOR methods. The three models were applied to over 1500 data points from field data and extrapolated and expanded using Eclipse Simulator. Analysis of the techniques by these methods show that surfactant injection achieved the greatest return in oil production with an average oil rate of 27.27Mstb per day. Polymer Injection produced the second greatest oil generation per day with an average rate of 25.19 Mstb per day, with water injection generating the least amount of oil at 25 Mstb per day. These results suggest that surfactant injection could be a promising method for enhancing oil recovery and extending the lifespan of existing oil and gas reserves.