Estimate Permeability from Nuclear Magnetic Resonance Measurements Using Improved Artificial Neural Network Based on Genetic Algorithm

Knowledge of the permeability distribution is critical to a successful reservoir model. Nuclear Magnetic Resonance (NMR) measurements can be used for permeability prediction because the T2 relaxation time is proportional to pore size. Due to the conventional estimators have difficult and complex problems in simulating the relationship between permeability and NMR measurements, an intelligent technique using artificial neural network and genetic algorithm to estimate permeability from NMR measurements is developed. Neural network is used as a nonlinear regression method to develop transformation between the permeability and NMR measurements. Genetic algorithm is used for selecting the best parameters and initial value for the neural network, which solved two major problems of the network: local minima and parameter selection depend on experience. Information gain principle is introduced to select the neural network's input parameters automatically from data. The technique is demonstrated with an application to the well data in Northeast China. The results show that the refined technique make more accurate and reliable reservoir permeability estimation compared with conventional methods. This intelligent technique can be utilized a powerful tool for estimate permeability from NMR logs in oil and gas industry.

[1]  De-Shuang Huang.  The local minima-free condition of feedforward neural networks for outer-supervised learning , 1998, IEEE Trans. Syst. Man Cybern. Part B.

[2]  I. O. Sikiru,et al.  Permeability Prediction Using Genetic Unit Averages of Flow Zone Indicators (FZIs) and Neural Networks , 2005 .

[3]  R. Ehrlich,et al.  Pore-Size Distribution And NMR In Microporous Cherty Sandstones , 1989 .

[4]  Shahab D. Mohaghegh,et al.  Virtual-Intelligence Applications in Petroleum Engineering: Part 2—Evolutionary Computing , 2000 .

[5]  W. E. Kenyon,et al.  Nuclear magnetic resonance as a petrophysical measurement , 1992 .

[6]  J. Robinson,et al.  Relations Between Pore Size Fluid and Matrix Properties, and NML Measurements , 1970 .

[7]  Alex Berson,et al.  Building Data Mining Applications for CRM , 1999 .

[8]  Shahab D. Mohaghegh,et al.  Virtual-Intelligence Applications in Petroleum Engineering: Part 1—Artificial Neural Networks , 2000 .

[9]  W. E. Kenyon,et al.  A Three-Part Study of NMR Longitudinal Relaxation Properties of Water-Saturated Sandstones , 1988 .

[10]  Sang Joon Kim,et al.  A Mathematical Theory of Communication , 2006 .

[11]  George R. Coates,et al.  A New Approach To Improved Log-Derived Permeability , 1973 .

[12]  Schwartz,et al.  Magnetic resonance as a probe of permeability in porous media. , 1987, Physical review letters.

[13]  A. Timur.  An Investigation Of Permeability, Porosity, And Residual Water Saturation Relationships , 1968 .

[14]  Yves Chauvin,et al.  Backpropagation: theory, architectures, and applications , 1995 .

[15]  Chris Morriss,et al.  Core Analysis By Low-field Nmr , 1997 .

[16]  W. E. Kenyon,et al.  Petrophysical Principles of Applications of NMR Logging , 1997 .