History Matching of Structurally Complex Reservoirs Using Discrete Space Optimization Method.

@inproceedings{P09_Suzuki,
 abstract = { History matching of structurally complex reservoirs is one of the most challenging tasks in reservoir
characterization, where the difficulty is attributed to the potentially large uncertainty in reservoir
geometry resulted from the limited quality of seismic data. In addition to the uncertainty in the
fault/horizon positions due to the low resolution of seismic images, structural interpretations or migration
results themselves are not unique and often rely on the subjective decision of an expert. In such
reservoirs, a traditional history matching approach performed by fixing the reservoir geometry to a single
interpretation may fail to match past production or may lead to future development planning based on a
‘wrong’ structural interpretation, since in many cases reservoir geometry has stronger impact on
production behavior than petrophysical properties distribution.
This paper proposes a new automatic history matching method for modeling a reservoir structure
considering both the large scale uncertainty related to structural interpretation and the smaller scale
uncertainty due to the error in fault/horizon positioning. The idea is to search for a history matched model
from the large set of prior structural model realizations, where the model realizations are created by
stochastically perturbing fault/horizon surfaces, using JACTATM/GOCAD and the method proposed by
Zhang and Caumon (2006), from multiple structural interpretation results. The multiple structural
interpretation results, which exhibit various numbers of faults, are provided for the fault/horizon
perturbation. The challenge is to solve an inverse problem where the (discrete) choice of the structural
interpretation is one of the parameters. Gradient-based methods do not apply in such inherently discrete
parameter space. Therefore, we introduce a new parameter space defined by a 'similarity distance', i.e. a
distance function that measures the similarity of geometry between any two model realizations from the
prior model space. We show that the production response in the space defined by a suitable similarity
measure is structured, i.e. not random, hence optimization/history matching with stochastic search
methods is effective. Using synthetic but realistic reservoir examples, we demonstrate that multiple
history matched models are found with reasonable CPU cost by applying either the neighborhood
algorithm (Sambridge, 1999) or the tree search optimization using geometric near-neighbor access tree
(Brin, 1995). },
 author = { Suzuki, Satomi AND Caers, Jef AND Caumon, Guillaume },
 booktitle = { Proc. 26th Gocad Meeting, Nancy },
 title = { History Matching of Structurally Complex Reservoirs Using Discrete Space Optimization Method. },
 year = { 2006 }
}