Chemical floods such as surfactant and/or polymer floods are enhanced oil recovery techniques that have been proven at both laboratory and field scale to increase sweep and/or displacement efficiency. Even though the compatibility and the efficiency of the injected chemicals are thoroughly tested and validated in the laboratory, uncertainty still remains regarding their actual performance in the reservoir. These uncertainties can result from the differences in the scale of investigation (core scale to field scale), lack of understanding of reservoir heterogeneity and connectivity, and the long term chemical performance and the chemical slug integrity in the reservoir.
In order to properly design a chemical flood, the uncertainties related to chemical performance have to be considered along with subsurface uncertainties during formal analysis using Design of Experiments. In addition to traditional subsurface uncertainty analysis, this work focused on two methodologies to systematically incorporate in-situ chemical performance uncertainties. The first methodology varies chemical properties such as polymer rheology, adsorption, permeability reduction, inaccessible pore volume and non-Newtonian behavior individually during uncertainty analysis. The drawbacks of this approach are the computationally cumbersome parameter combinations and an increased possibility of testing unphysical combinations of chemical parameters.
The second methodology considers the overall chemical performance in the reservoir as an uncertainty parameter. In this case, the chemical properties are not varied individually but as a set to express the uncertainty range of chemical quality. The advantage of using chemical performance as an uncertainty parameter is that the possibility of testing unphysical combination of chemical properties is eliminated and there are less uncertainty parameters (thus increased computational efficiency). The drawback of this approach is that inadvertent bias can be introduced in the uncertainty analysis because of limited knowledge of the subsurface, and extreme end-points in the uncertainty space may remain untested.
The two approaches are compared in this paper and case studies were conducted. The results show how realistic low, mid and high performance models were generated that account for both subsurface and chemical uncertainties.