Treating Uncertainties in Reservoir Performance Prediction with Neural Networks

Abstract

Abstract In development projects reservoir parameters are only known within certain ranges – a fact that allows various realisations of the subsurface. Because of the computational time involved, not all of the possible parameter combinations can be covered by simulation models to obtain a probability distribution of possible outcomes. Creating a response surface that is based on a reduced number of simulation runs becomes necessary. Such a response surface can be utilized to approximate results for numerous different variations of input parameters. In contrast to the widely used methodology of fitting a polynomial model to the results of a limited number of simulation runs, an approach, where reservoir response is captured by an Artificial Neural Network (ANN) has been investigated. As a first step the most sensitive parameters which affect the performance of the simulation model are determined with a limited number of model runs. These simulation runs cannot cover each of the possible model realisations, but they are intended to span over the whole range of input parameter variations. By training an ANN on the so gained simulation results, a model, which is able to interpolate between the individual simulation scenarios is created. In this way a large variety of realisations can be approximated by a limited number of reservoir simulation models. The trained ANN model is used in Monte Carlo Simulation to generate the probability distribution of all possible outcomes. Due to the very low CPU consumption of the ANN, a large number of realisations can be calculated in a short amount of time.