Evaluation of the impact of reservoir heterogeneity and mixing on low salinity waterflooding

Abstract

This study investigated the impact of reservoir heterogeneity and physical diffusion/dispersion on low salinity waterflooding (LSW) using detailed numerical simulation. Two different methods for modelling the impact of LSW on relative permeability and hence on recovery were investigated: the salinity threshold model (ECLIPSE and STARS) and the ion exchange model (GEM). These simulators were validated by comparing their predictions with the analytical solution for immediate LSW. Physical diffusion/dispersion were characterized using the dimensionless transverse dispersion number (NTD) which was shown to be a robust measure for determining their impacts on the performance of LSW. The impact of numerical diffusion was evaluated for a range of grid resolutions and a correlation developed for estimating the maximum physical longitudinal dispersion that can be captured for a given grid resolution. The impact of transverse numerical diffusion was found to be very small. On laboratory scales, it was found that diffusion improves the oil recovery in layered high net-to-gross models but reduces the effectiveness of continuous LSW in models containing a thick shale filled with connate water. On the reservoir scale, the effect of dispersion on LSW was found to be minor compared with reservoir heterogeneity. The adverse effect of an aquifer was found to be most significant during the production of the connate water bank. If low salinity water is injected immediately as a slug, it was found that the best slug size is around 0.4-0.6 PV assuming a total water injection of 1 PV. If it is injected after a conventional water flood then the slug size increases to 1.3 PV assuming a total water injection of 2 PV. Finally, an analytical pseudoization method was proposed to upscale LSW by modifying Hearn’s method for layered reservoirs to include the connate water bank. This method was shown to give improved predictions over those obtained by the original Hearn’s method when compared using a 2D layered model.