Updating Uncertainty in the Conceptual Geological Representation of Fractured Reservoirs Using Production Data

Fractured reservoirs are characterized by potentially large uncertainty in their geological descriptions. An additional difficulty in modeling such reservoirs is that the detailed geological descriptions of discrete fracture networks are difficult to flow simulate with current commercial simulators. As a consequence, much of the geological information and its uncertainty are lost at the stage of history matching. In this paper we propose an approach whereby not the reservoir models, but the geological conceptual descriptions are probabilistically updated when new production is acquired. To achieve this, we design a geo-modeling workflow that turns conceptual DFN description into multiple alternative training images for dual-medium flow purposes. Multiple-point geostatistics can then rapidly generate reservoir model for dual-medium flow simulation. Since uncertainty is large, the set of possible training images can be very large. To reduce this set, we use a pattern-based clustering method to select a limited amount of training images representing uncertainty of the entire set. The production data is then used 1) to eliminate training images inconsistent with new production and 2) calculate updated likelihood of the remaining one using Bayes’ rule. We illustrate our workflow on a middle-Eastern fractured reservoir.