Impacts of reservoir heterogeneities in the form of shale barriers on SAGD production can be analyzed by generating a large number of realizations of shale barrier configurations and subjecting them to flow simulation. However, visualizing and quantifying the (dis)similarities among these realizations is often challenging. A workflow that applies multidimensional scaling (MDS) and cluster analysis techniques is developed to represent the uncertain influences of different shale barrier configurations on SAGD production and to quantify the dissimilarities between realizations.
A two-dimensional homogeneous simulation model is employed, and reservoir heterogeneities are simulated by superimposing sets of idealized shale barriers on the homogeneous model. The petrophysical properties, such as the porosity, permeability, initial oil saturation and net pay thickness, have been taken from average values for several pads in Suncor's Firebag project. One thousand models with various shale barrier configurations are then subjected to flow simulation to estimate SAGD production in each case. First, a distance function, which measures the dissimilarity in production responses between any two given shale barrier configurations, is formulated. Next, MDS maps the resultant distance matrix into an n-dimensional Euclidean space, where k-means clustering technique is applied to group the models into multiple clusters. Although the precise distribution of shale barriers would vary among models within the same cluster, it is expected that their impacts on SAGD production are similar.
Specific features corresponding to the shale barriers in each cluster are analyzed, and they are studied to infer any potential correlation between SAGD production and the particular shale distribution characteristics. The results are employed to revise the original set of realizations by adding new models to clusters with fewer members and removing models from clusters with redundant members. The new models are subjected to flow simulation to verify their membership to the assigned clusters, and good agreement in the results has been observed.
Data-driven or AI-based modeling approaches for production analysis have gained much attention over recent years. In most cases, a training data set consisting of many different realizations of reservoir heterogeneity is needed. A key question remains: "how many realizations are needed to span the model parameter space?" The proposed workflow offers an efficient and systematic method for constructing data sets that maximize the spanning of the model parameter space, without exhaustively sampling similar realizations and subjecting them to flow simulation. This is a particularly important consideration when 3D models are utilized. Furthermore, the ability to visualize and select representative models or scenarios from individual clusters has important potential for facilitating improvements in operations design in the presence of reservoir heterogeneities.