Determination of miscible CO2 flooding analogue projects with machine learning

Abstract

Using analogue reservoirs for comparison and benchmarking is a comprehensive method for ensuring accurate measurement and gaining a better understanding of potential oil recovery enhancement prospects. The approach of leveraging information from analogue reservoirs is extended to CO 2 EOR project management in this study. We present a novel application of machine learning clustering algorithms to the rapid identification of analogues for new projects without having to sift through massive amounts of data. We use machine learning clustering methods to group successfully executed miscible CO 2 flooding projects into clusters of projects with similar fluid/reservoir characteristics and to identify analogues for new target projects. Porosity, permeability, oil gravity and viscosity, reservoir pressure and temperature, minimum miscibility pressure (MMP), and depth were all input parameters. Data from nearly 200 miscible CO 2 EOR projects around the world were clustered using the Agglomerative Hierarchical Clustering Algorithm (HCA), K-Means, and K-Median techniques. To reduce information redundancy in high-dimensional data, Principal Component Analysis was used as a pre-processing step. Three evaluation indices (the Davies Bouldin, Calinski Harabasz, and Silhouette Coefficient Scores) were used to compare the efficiency of the clustering algorithms and choose the best one for this dataset. A Principal Component – weighted Euclidean distance similarity metric was computed using three existing miscible CO 2 flooding projects (Weyburn, Hansford Marmaton, and Paradis) as test cases to confirm the clustering results. The clustering analysis identified five different classes of miscible CO 2 projects, each with its reservoir and fluid characteristics. Type 1 projects, in general, are those that are carried out primarily in shallow carbonate reservoirs at the lowest temperatures and pressures of any database project with typical porosity and permeability. More than 60% of Type 2 projects are in sandstone reservoirs. They are at shallower depths with lower temperatures, pressures, porosities, and permeabilities than project Type 4. Type 3 projects are typically undertaken in carbonate reservoirs with medium depths and temperatures but the highest reservoir pressures. This project type has medium porosity and permeability. In comparison to project Type 2, Type 4 primarily consists of projects conducted in sandstone formations at great depths with high reservoir temperatures and pressures. The porosity and permeability of these project types are average. Finally, Type 5 projects are typically undertaken in sandstone formations at average depths, temperatures, and pressures, but with the greatest porosity and permeability of all project types. In addition to the clustering analysis, the distance similarity metric used in this work identified projects that were most like the test miscible CO 2 flooding project cases. Key rock and fluid properties, well types, and best infill drilling strategies, recovery improvement strategies, and production performance can all be learned from identified analogue projects. This data can be used to improve the operational, technical, field, and well-planning decisions for new CO 2 flooding projects. The workflow demonstrated in this paper is easily adaptable to data sets from other flooding projects. • K-Means clustering model can be used effectively to categorize miscible CO 2 flooding projects. • The global miscible CO 2 flooding projects in collated dataset can be broadly divided into five classes by K-means clustering model. • A calculated Principal Component – weighted Euclidean distance similarity metric can be used to successfully validate cluster groupings. • Ideal analogue projects with similar fluid and reservoir properties can be identified with K-Means and the distance-based metric