Linear vs. nonlinear porosity estimation of NMR oil reservoir data

Abstract

Nuclear magnetic resonance is widely used to assess oil reservoir properties especially those that can not be evaluated using conventional techniques. In this regard, porosity determination and the related estimation of the oil present play a very important role in assessing the eco1nomic value of the oil wells. Nuclear Magnetic Resonance data is usually fit to the sum of decaying exponentials. The resulting distribution; i.e. T2 distribution; is directly related to porosity determination. In this work, three reservoir core samples (Tight Sandstone and two Carbonate samples) were analyzed. Linear Least Square method (LLS) and non-linear least square fitting using Levenberg-Marquardt method were used to calculate the T2 distribution and the resulting incremental porosity.

Parametric analysis for the two methods was performed to evaluate the impact of number of exponentials, and effect of the regularization parameter (?) on the smoothing of the solution. Effect of the type of solution on porosity determination was carried out. It was found that 12 exponentials is the optimum number of exponentials for both the linear and nonlinear solutions. In the mean time, it was shown that the linear solution begins to be smooth at α = 0.5 which corresponds to the standard industrial value for the regularization parameter. The order of magnitude of time needed for the linear solution is in the range of few minutes while it is in the range of few hours for the nonlinear solution. Regardless of the fact that small differences exist between the linear and nonlinear solutions, these small values make an appreciable difference in porosity. The nonlinear solution predicts 12% less porosity for the tight sandstone sample and 4.5 % and 13 % more porosity in the two carbonate samples respectively.