Integration of data-driven models for dynamic prediction of the SAGD production performance with field data

Abstract

Steam-assisted gravity drainage (SAGD) is a successful thermal recovery process and has been widely applied to oil sands production. The prediction of oil production of the SAGD process plays a significant role in decision-making, where numerical simulation is one of the tools which support engineers. However, the traditional numerical simulation process for the SAGD production prediction, such as history matching, sensitivity analysis, and predictive runs, is a time-consuming process that is always associated with heavy computational costs. Engineers require reliable alternative modeling tools for the management of SAGD production. In this study, a data-driven model-based workflow is formulated and tested for the prediction of the SAGD production performance. After a series of data processing and integration of real field data, the machine learning algorithms are successfully utilized to predict future production performance based on past production information and operational conditions. A comparison of different machine learning algorithms shows that a Gated Recurrent Unit (GRU) based model has the best predictive ability.