Permeability Prediction from Clay Volume Using Artificial Intelligence

Abstract

Abstract The importance of permeability can be observed by the massive number of studies on its prediction as reported in literatures. Common methods of permeability prediction depend mainly on core analysis and pressure test data in the lab. However, there are difficulties associated with predicting permeability from traditional methods. This motivates the search for new methods. In this study, the prediction of reservoir permeability from clay volume by using Artificial Intelligence is proposed as an innovative solution. One of these AI methods is Artificial Neural Networks. The ANN approach is implemented in this paper in order to predict reservoir permeability. This study used the volume of clay and porosity as input parameters. The model data was mainly obtained from well logs and validated by core data from 21CXRM Palaeozoic project in the North Sea. The training of the ANN model was completed using the Backpropagation Neural Network (BPNN). More than 25,000 data points were used in the development of the ANN model. The data was divided randomly to train and test the ANN model. The results of this study illustrate excellent correlation between the actual permeability values and the predicted values from the ANN model, i.e., 0.9985 and 0.9991 for the training group and the testing group, respectively. The ANN successfully predicted the permeability for a wide range of reservoir permeability values at different reservoir depths. A mathematical equation was developed to demonstrate the relationship between the used input data and the output results. This proves the capability of AI techniques to estimate reservoir parameters in general.