Screening enhanced oil recovery methods using the machine-learning algorithm K Nearest Neighbors (KNN)

Abstract

The process of selecting enhanced oil recovery methods, known as EOR (Enhanced Oil Recovery) methods, has been a great challenge within the oil and gas industry because it depends on the characteristics of the reservoir and economic aspects. This has usually been done through the use of comparative tables, where the application of each of these methods is determined by specific ranges of reservoir parameters. However, this selection methodology may generate biased results due to its qualitative nature and interpretation depending on the criteria of the professional in charge of the process, therefore, it is pertinent to seek the implementation of techniques that provide better results. Therefore, this article focuses on improving the accuracy of the selection of these methods, implementing data-driven decisions through a machine-learning model, specifically the Nearest K model, whose objective is to predict which enhanced oil recovery method is the most adequate to implement in an oil well located in the Amazon region of Ecuador, whose reservoir contains heavy oil. It is worth mentioning that the algorithm was trained with more than 200 enhanced oil recovery projects from several oil-producing countries, where the input data of the model are seven reservoir parameters, which are porosity, permeability, depth, temperature, API gravity, viscosity, and oil saturation whereas the output parameters of the model are 5 enhanced recovery methods, which are steam injection, CO2 injection, hydrocarbon injection, polymer injection, and in-situ combustion. The Model was successfully trained because its metrics yielded acceptable values, as its accuracy was about 84%, the error rate equal to 0.16, and as a result, it predicted that the most suitable EOR method to implement into this oil well is the steam injection method.