The use of Ann for the prediction of the modified relative permeability functions in stratified reservoirs

Abstract

The paper presents a method of instantaneous construction of relative permeability pseudo functions in analytical form upscaled to a coarser computational grid using a system of artificial neural networks. The coefficients of these functions can be forecasted by the neural network. The learning dataset is based on a preliminary series of calculations at the reference values of the system parameters the exponents of the initial functions, the liquid phases viscosity ratio, the statistical parameters of distribution laws of the reservoir’s properties. The latter may be obtained according to the primary well logging data with no need for building a detailed geological model.