Relative permeability plays an important role in the characterization of multiphase flow in porous media. Such data are used as an input parameter for reservoir simulation models to determine the phase distribution, residual saturation, and predict the future performance of the reservoir. The inaccurate characterization of the relative permeability results in false representations of the velocity fields in the reservoir, inaccurate forecasts of production levels, and leads to wrong decisions, such as the incorrect selection of the well locations or the best recovery technique to be implemented. Besides, in simulations with a large number of data, as is the case of reservoirs with several grid blocks, the computational time for processing can last for several days. Given these difficulties, this work presents a methodology for forecasting relative permeability curves using feedforward artificial neural networks (ANNs) to improve simulation models in a short period of time. The input variables consisted of history data from production and injection wells and the analysis was based on the correlation between these data and relative permeability curves. Such methodology proved to be an alternative as a method of history matching, since the networks presented low error, with values very close or equal to the historical values, leading to a better production forecast with less computational demand.
Keywords: Artificial Neural Network. History Matching. Relative Permeability