Application of Artificial Neural Network to Estimate Permeability From Nuclear Magnetic Resonance Log

摘要

Many oil reservoirs have heterogeneity in rock properties. Understanding the form and spatial distribution of these heterogeneities is fundamental to the success of reservoir description. Permeability is one of the fundamental rock properties to characterize flow potentials of reservoir when subjected to applied pressure gradients. A number of mathematical models have been suggested in the literature to simulate and quantify this property. However, common to them is difficulties in being able to model appropriately various geological variations associated with any reservoir. This study explores the benefits of using artificial neural network and NMR log in the permeability predictive model development. ANN was used to capture the non- linearity issues between the dependent and independent variables while the transversal relaxation times (T2) data is capable of capturing intrinsic rock properties at pore scale level. Out of verified datasets available, 60% of the datasets were used for the training process and the remainder for testing and validation. The input data was transverse relaxation time data, its mean value of the bins, mean square value of the bins and the maximum value of the bins. The developed ANN was trained, tested and validated using MATLAB Neural network toolbox trained with backward propagation scheme. The result shows a very good performance of ANN when compared with other existing empirical correlations adopted in this study.