Modeling rock permeability from NMR relaxation data by PLS regression

This study explores the application of the partial least squares regression (PLSR) technique to rock permeability prediction from nuclear magnetic resonance (NMR) relaxation data. A total of 68 Brazilian sandstone cores selected from reservoirs and outcrop analogs were fully saturated and analyzed by NMR. The permeability of the cores ranged from 0.007 to 9,800 mD. From their 1H transverse relaxation times (T2) measured at 2 MHz, two PLSR models were developed for the relaxation spectra and the raw relaxation curves. Both models led to more uniform and accurate predictions (RMSE = 0.47 and 0.50 log mD, respectively) compared with the classical Kenyon model (RMSE = 0.78 log mD).

► We evaluated the application of partial least squares regression (PLSR) in rock permeability prediction from NMR data.
► We used both relaxation spectra and the raw 1H NMR relaxation curves.
► Both PLSR models lead to more uniform and accurate predictions when compared to the classical Kenyon model.