Toward evaluation and screening of the enhanced oil recovery scenarios for low permeability reservoirs using statistical and machine learning techniques

Abstract

The concurrence of the oil demand increment and running out of fossil fuels have brought about special attention toward the tight and low permeability reservoirs. The decision-making regarding the most appropriate EOR approach suiting these target zones is not straightforward and may be associated with more sophistication than the conventional high permeability alternatives. This fact justifies the attempt to take a closer look at the outcomes of the EOR scenarios that have been implemented in the analogous oilfields. This is where the current study lies. This piece of study provides a rich databank with the hope to shed light on the potential EOR scenarios for the onshore/offshore low permeability reservoirs. The databank that embeds over 320 projects from about 218 oilfields situated in 16 worldwide countries is manipulated for two purposes: (1) evaluating the IOR projects in the worldwide low permeability onshore/offshore oilfields and (2) initiating various EOR screening tools based on the statistical plots, ensemble learning algorithms of Random Forest (RF), Gradient boosting Machine (GBM), and Extreme Gradient boosting (XGBoost), and the evolutionary algorithm of Gene Expression Programming (GEP). Comprehensive investigation and statistical analysis are conducted on the range of important parameters that can affect the EOR screening in this databank. Seven screening criteria of permeability, porosity, API, viscosity, lithology, depth, and temperature were considered for this purpose. The evaluation has demonstrated that the gas-based scenarios have the most contribution to the EOR projects of the world's onshore and offshore oilfields, respectively. Furthermore, the generated EOR screening tools have established themselves as valuable tools for the scope of this study. Among the AI-based approaches, the RF outperformed the alternatives with the training and test accuracies of 0.99 and 0.90, respectively. In the end, the achieved knowledge and the developed screening tools are used in a case study for elucidating the potential EOR scenarios for a given oilfield.