Reservoir Characterization Using Well Logs and Digitized Core Images: A Case Study from the Montney Formation

Abstract

Tight reservoirs typically have low permeability (<0.1 mD) and small scale (mms-cm) heterogeneity, which are difficult to characterize using conventional methods. The aim of this study is to develop a reliable permeability predictor and to characterize the small scale heterogeneity based on a limited petrophysical dataset from the finely laminated tight gas formation, the Montney Formation. Three approaches, multiple variable regression, empirical model and artificial neural networks (ANNs), have been used to predict permeability based on well logs. The ANN developed from the workflow produces predictions for permeability with R2 = 0.99 and mean square errors equal to 0.0069, which is more accurate than the other two methods. A digital image analysis program is developed in Matlab, which can correct the effects of fractures, converts the image into gray value data and identifies the locations and thicknesses of laminations. By analyzing the digitized image data, lamination scale heterogeneity can be quantitatively characterized. By incorporating the gray values into the ANN, the network’s ability to generalize is much improved, increasing the test set R2 value from 0.26 to 0.89.