Tight oil reservoirs are mainly developed by water injection. The conventional method could not effectively characterize the sand body connectivity between oil and water wells since it only uses static geological parameters or dynamic production data to identify the connectivity between wells. In this paper, a novel machine learning evaluation method for inter-well sand body connectivity using both static geological and dynamic production data is constructed. This method is based on the CatBoost algorithm. Three types of sand body connectivity are classified by analysis of four geological factors including porosity, permeability, shale volume and net-to-gross ratio, and three dynamic production factors including oil production total (OPT), liquid production total (LPT) and water injection total (WIT). Finally, the novel method proposed in this paper is applied to predict the sand body connectivity between oil and water wells using the data from a tight oil reservoir located in the Ordos Basin, China. The results show that the proposed method can improve the forecast accuracy of inter-well sand body connectivity from 50% to 85%.
Keywords Tight oil reservoir, Sand body connectivity, CatBoost model, Machine Learning Evaluation Method, Parameters optimization