Flow field characterization and evaluation method based on unsupervised machine learning

Abstract

Long-term water flooding can be inefficient due to imbalanced displacement, resulting in sub-optimal recoveries. Therefore, the quantitative characterization and evaluation method of the flow field is significant for further development and adjustment. In this study, based on streamline simulation results, decoding rules were formulated to extract streamline data. Considering the attribute feature of each node along the streamline, flow field characterization indices were constructed. And combining the fast density peaks clustering (FDPC) algorithm and silhouette coefficient (SC), a novel flow field evaluation method based on streamline clustering was proposed. Results indicated that the invalid water circulation channels and areas with developing potential in reservoirs can be identified. Besides, the distribution of flow field can also be observed, and the characteristics of the identified region can be analyzed, which provides a basis for the subsequent flow field adjustment. Meanwhile, streamlines between injection and production wells can be subdivided to describe the distribution of water displacement capacity. Based on the evaluation results of streamline clustering, Gudao oilfield adjusted the flow field through injection-production adjustment, which effectively improved the development effect of the block and increased the recovery degree from 50.3% to 52.4%. This work provides an effective evaluation method to directly characterize the displacement changes for underground flow field, and it is of great significance for the decision-making of water flooding structure adjustment and optimization.