Optimization-Based Reservoir Modeling: Example On A Million-Cell Model

3-D reservoir characterization procedures aim at achieving a spatial distribution of reservoir properties that is<br>consistent with the available reservoir data. Geostatistical algorithms, probably the most popular approach for<br>reservoir characterization, have indeed provided a practical methodology to this problem. These techniques,<br>however, have been somewhat limited on the number of data sets they can account for and on the propagation<br>of data uncertainties and data resolution on reservoir properties to the final model.<br>We have proposed an optimization-based approach for reservoir modeling (Gouveia et al., 1998) that can<br>overcome some of these limitations. This algorithm provides a framework to integrate a broad spectrum of data<br>sets (well, seismic, production and geological data) in such a way that the respective degrees of data<br>uncertainty and resolution are taken into consideration. These features come at a considerable computational<br>cost when compared to geostatistical techniques. However, via a modified Monte Carlo sampling procedure, we<br>were able to reduce the computational cost to the point that the proposed methodology can be applicable to<br>more realistic reservoir modeling situations. Here, we report results that illustrate the performance of the<br>optimization algorithm on the modeling of a synthetic reservoir parameterized by a one-million-cell model<br>defined on a Cartesian grid.