Abstract

Current computer technology has given us the ability to generate realizations of spatial reservoir properties as never before. However, achieving the goal of generating reservoir descriptions which are consistent with all of the available information continues to be a difficult and time-consuming process. Especially, the inability to constrain probabilistic-based reservoir descriptions using (dynamic) production data is becoming recognized as one of the primary short-comings of existing techniques.

This paper describes a procedure to identify, quantify and incorporate within a conditional simulation procedure the spatial reservoir characteristics which dominate well performance. Primary production data are used to determine controlling spatial characteristics. The primary production parameters are combined with waterflood constraints and the method of simulated annealing to produce realistic reservoir descriptions––both in terms of their spatial characteristics and simulated well performance. Although excellent matches of well performance are also obtained using these "indirect performance constraints" while excluding the variogram constraint, it is shown that the resulting alternative reservoir descriptions poorly reproduce actual spatial reservoir properties. Thus, it is concluded that spatial correlation structures cannot be extracted from performance data. The robustness and flexibility of the approach is demonstrated using a two-dimensional areal, field scale reservoir study. The proposed technique can be used to greatly reduce the uncertainty of predicting future reservoir performance.

Introduction

A probabilistic modeling approach is suitable for translating geologic uncertainties to uncertainties in future reservoir performance. Integration of all available data into the reservoir description minimizes the range of uncertainty. The concept of minimizing uncertainty by including additional conditional simulation constraints is illustrated in Figure 1. Stochastic modeling is used to generate several equally probable reservoir descriptions. These reservoir descriptions are then flow simulated to determine the range of possible outcomes. Frequency distributions of an example flow simulation output parameter, ultimate recovery, is depicted in Figure 1. If an insufficient number of relevant constraints have been imposed on the system, it is very likely that the actual ultimate recovery represents an unlikely occurrence. This situation is represented by the two frequency distributions labelled "conventional constraints." Note that depending on the inadequacies of the imposed constraints, the resulting distributions can be unrealistically broad or narrow. However, if additional constraints can be imposed, each having a significant influence on fluid flow characteristics, then the corresponding frequency distribution of outcome parameters should be greatly reduced. This is illustrated in Figure 1 by the distribution labelled "relevant constraints." Not only should the spread in possible outcomes be reduced but, as indicated by this second scenario, the actual value of ultimate recovery should fall well within the bounds of the revised frequency distribution. A more informed reservoir management decision can be made given the reduced range in possible outcomes.

Thus, it becomes important to identify the spatial reservoir features which have the greatest impact on reservoir performance. Once these features have been identified, a method is required to quantify these characteristics and incorporate them as constraints within a probabilistic method. This basic approach was used previously for investigating permeability heterogeneity constraints for a five-spot waterflood pattern. The work presented here, supplemental to the previous investigation, focuses on primary production and integration of both primary- and secondary-based constraints into one conditional simulation method.

P. 749^

This content is only available via PDF.
You can access this article if you purchase or spend a download.