Permeability Prediction from Well Log Data Using Artificial Neural Networks: A Case Study of the Hawaz Formation in D-Field, Libya

Abstract

Permeability is an important parameter in the characterization of any hydrocarbon reservoir. Yet, despite its vital importance, it is one of the most difficult and controversial petrophysical properties to calculate accurately due to nonlinearity and uncertainty in the dataset. Formation permeability is often measured in the laboratory from cores or evaluated from well test data. Core analysis and well test data, however, are only available from a few wells in a field, while the majority of wells are logged. The aim of this paper is to design an artificial neural network (ANN) model to predict formation permeability using a dataset from the Hawaz Formation in the D-field NC-186 concession, East Murzuq Basin, Libya. In this study, a back-propagation neural network (BP-ANN) model was built using the Python programming language. A total of 950 core horizontal permeability measurements and their corresponding well log data were collected from four wells to build the model. Traditional statistical analysis of the porosity/permeability relationship in cored well data yielded no reliable correlations for predicting permeability in uncored wells with R² less than 15%. A supervised BP-ANN model was trained successfully and was successfully able to predict the permeability of the formations. Despite the presence of high reservoir heterogeneity, the permeability profile predicted by the ANN model using well log data agrees well with core permeability, which clarifies the applicability of this method.