A New Algorithm Model Based on Extended Kalman Filter for Predicting Inter-Well Connectivity

Abstract

Given that more and more oil reservoirs are reaching the high water cut stage during water flooding, the construction of an advanced algorithmic model for identifying inter-well connectivity is crucial to improve oil recovery and extend the oilfield service life cycle. This study proposes a state variable-based dynamic capacitance (SV-DC) model that integrates artificial intelligence techniques with dynamic data and geological features to more accurately identify inter-well connectivity and its evolution. A comprehensive sensitivity analysis was performed on single-well pairs and multi-well groups regarding the permeability amplitude, the width of the high permeable channel, change, and lasting period of injection pressure. In addition, the production performance of multi-well groups, especially the development of ineffective circulation channels and their effects on reservoir development, are studied in-depth. The results show that higher permeability, wider permeable channels, and longer injection pressure maintenance can significantly enhance inter-well connectivity coefficients and reduce time-lag coefficients. Inter-well connectivity in multi-well systems is significantly affected by well-group configuration and inter-well interference effects. Based on the simulation results, the evaluation index of ineffective circulation channels is proposed and applied to dozens of well groups. These identified ineffective circulation channel changing patterns provide an important basis for optimizing oil fields’ injection and production strategies through data-driven insights and contribute to improving oil recovery. The integration of artificial intelligence enhances the ability to analyze complex datasets, allowing for more precise adjustments in field operations. This paper’s research ideas and findings can be confidently extended to other engineering scenarios, such as geothermal development and carbon dioxide storage, where AI-based models can further refine and optimize resource management and operational strategies.