History Matching and Uncertainty Quantification of Fractured Reservoirs by Updating
DFN Parameters using Data Assimilation Workflow
Siavash Nejadi, Japan Trivedi*, Juliana Leung
nejadi@ualberta.ca, jtrivedi@ualberta.ca, juliana2@ualberta.ca
Civil and Environmental Engineering Department, School of Petroleum, University of Alberta
3-122 Markin/CNRL Natural Resources Engineering Facility
Edmonton, Alberta, Canada T6G 2W2
Fractured reservoirs are highly heterogeneous and can be characterized by probability
distributions of fracture properties in a discrete fracture network (DFN) model. The relationship
between production performance and the fracture parameters is vastly non-linear, rendering the
process of adjusting model parameters to match both static geologic and the dynamic production
data challenging.
This paper presents an integrated approach of data assimilation for history matching and
characterization of fractured reservoirs by directly updating DFN parameters. This new
methodology includes generating multiple DFN realizations, upscaling the models for numerical
multiphase flow simulation, and updating the DFN model parameters using dynamic flow
responses such as continuous rate and pressure measurements. The available geologic and
tectonic information such as well-logs, seismic, and structural maps are incorporated in the
FRACMAN® software to infer probability distributions of relevant fracture parameters
(including aperture, length, connectivity, and intensity) and to generate multiple DFN
realizations. The DFN models are further upscaled into an equivalent continuum dual-porosity
model in the software using either analytical approaches (Oda) or the dynamic methods. The
upscaled models are subjected to the flow simulation and their production performances are
compared to the true recorded responses. An automated data assimilation algorithm is
implemented to reduce the uncertainties of the DFN models. Components of vectors representing
the principal flow directions and average fracture orientations are obtained by means of
eigenvector decomposition of the permeability tensor and optimized in the algorithm. In
addition, both, global fracture intensity and the local grid based P32, which highly affects the
fluid flow pattern and rate in different regions of the reservoir, are updated. At the same time,
dynamic localization techniques are applied to improve the quality of history matching results
and minimize spurious long-distance correlations.
Two case studies with various fractures sets are presented. In each case, the initial realizations
were generated by means of Monte Carlo simulations, using the observed P10 information at the
well locations. Fracture intensity, orientation and conductivity of different fracture sets were the
uncertain parameters in our studies. Using the proposed methodology, parameters of different
fracture sets were satisfactorily updated. Implementation of this data assimilation approach
resulted in multiple equally probable DFN models and their upscaled flow simulation models,
which honor the geological information, and at the same time they match the dynamic
production history.