Application of machine learning algorithms in classification the flow units of the Kazhdumi reservoir in one of the oil fields in southwest of Iran

Abstract

By determining the hydraulic flow units (HFUs) in the reservoir rock and examining the distribution of porosity and permeability variables, it is possible to identify areas with suitable reservoir quality. In conventional methods, HFUs are determined using core data. This is while considering the non-continuity of the core data along the well, there is a great uncertainty in generalizing their results to the entire depth of the reservoir. Therefore, using related wireline logs as continuous data and using artificial intelligence methods can be an acceptable alternative. In this study, first, the number of HFUs was determined using conventional methods including Winland R35, flow zone index, discrete rock type and k-means. After that, by using petrophysical logs and using machine learning algorithms including support vector machine (SVM), artificial neural network (ANN), LogitBoost (LB), random forest (RF), and logistic regression (LR), HFUs have been determined. The innovation of this article is the use of different intelligent methods in determining the HFUs and comparing these methods with each other in such a way that instead of using only two parameters of porosity and permeability, different data obtained from wireline logging are used. This increases the accuracy and speed of reaching the solution and is the main application of the methodology introduced in this study. Mentioned algorithms are compared with accuracy, and the results show that SVM, ANN, RF, LB, and LR with 90.46%, 88.12%, 91.87%, 94.84%, and 91.56% accuracy classified the HFUs respectively.