Identifying inter-well connectivity is essential for reservoir development adjustment and optimization, but the actual working and reservoir conditions are complex, making the task difficult to do manually. Therefore, a set of unsupervised machine learning algorithms based on affinity propagation (AP) is developed in this work to cluster and map production data of oil wells in two dimensions and then extract inter-well topology to achieve automatic identification of inter-well connectivity. To better respond to the conditional independence of the variables, the Graphical Lasso algorithm is used to find the inter-well correlation matrix. Finally, the Local Linear Embedding (LLE) algorithm is used to embed the production data of the wells into a two-dimensional plane to visualize the clustering results and inter-well connectivity relationships.

Results show that production wells close to the fault can be aggregated automatically, which proves that the method can identify the impermeable boundaries. In addition, the process can automatically cluster production wells of different permeability zones and distinguish production wells at the junction of low and high permeability zones. Finally, this method is applied to the production data of 63 wells in an actual reservoir. The model divides the oil well into four macro-level regions, which is consistent with geological understanding. At the microscopic level, five groups of wells with highly similar production variations were automatically detected, and the presence of high permeability channels between them was accurately identified. The proposed method has important practical significance for reservoir development adjustment and geological understanding.

You can access this article if you purchase or spend a download.