Assessment of Uncertainty in Reservoir Production Forecasts Using Upscaled Flow Models

Abstract

Production forecasts for a given reservoir, with associated uncertain-ties, must be based on a stochastic model of the reservoir coupled with a fluid flow simulator. These flow simulations may be computationally prohibitive if they are performed on the fine scale geostatistical model. To accelerate these computations, coarsened (or upscaled) reservoir models are generally simulated. The recovery predictions generated using the upscaled models will inevitably differ from those of the underlying fine scale models and may display severe biases and erroneous uncertainties. In this paper, a model that accounts for these biases and changes in uncertainties is described. Issues related to the selection of fine scale calibration runs and the performance of the method using different upscaling procedures are considered. Two test cases involving water injection are presented. These examples demonstrate the large errors that can result from standard upscaling techniques, the ability of the error modeling procedure to nonetheless provide accurate predic-tions (following appropriate calibration), and the benefits of using a more accurate upscaling procedure.