Permeability Estimation for Upper Shale Member in Southern Iraqi Oil Field Using Machine learning and Hydraulic Flow Units Methods

Abstract

Permeability estimation in the Upper Shale Member is rather complex due to the heterogeneity. Conventional logs do not provide permeability directly, but the core data does. Core operations and analysis processing are all costly. An alternative method is used by dividing the reservoir rock into groups (rock types) with similar characteristics in porosity and permeability for cored wells. These identified groups for the core wells can be applied to the uncored wells to predict their groups and permeabilities. Four wells were used in this study: one cored well and three uncored wells. Two classification methods of rock types were used; the first is the machine learning method (supervised and unsupervised), a new technics invention, and the second method is the hydraulic flow unit. Four groups were selected by using Machine Learning and hydraulic flow unit methods. A regression equation for each rock type was obtained to predict permeability. Then, establishing a relationship between the results of the core data groups and the logs' responses for the core well is referred to as "correlation," which is training the logs that provide an estimating model. A validation check was completed by comparing the rock type and permeability estimation from the core with the rock type and permeability estimation from the logs, which were highly matched. Finally, The rock type and permeability can be estimated for the uncored wells using only the logs that were learned from the generated training data set for the cored wells. Due to the use of more than one model to predict permeability, it was clear that one was more compatible than the other. The flow zone indicator, a classical method, still has a better rock type and permeability estimation than the new techniques, which use data mining and mathematical algorithms on a clustering basis.