Volume 51

Timing prediction model of CO2 channeling for intelligent early warning based on a transfer learning framework Beichen Zhao, Diwu Pengxiang, Yuedong Yao, Chaoxiang Wang, Zheyu Xiao, Hanze Liu, Yafei Wei

Abstract

CO2 flooding technology is a highly efficient method of enhanced oil recovery (EOR), but it can lead to CO2 channeling, which significantly reduces oil recovery and threatens well safety. Due to the complex geological structure of reservoirs, there is a high level of uncertainty. Traditional numerical simulation and empirical methods are limited in their ability to accurately predict CO2 channeling, resulting in uncertain channeling times. In addition, there are CO2 channeling intricately influenced by a multitude of factors, including fluid properties, inter-well connectivity, injection-production mechanisms, and CO2 channeling capacity, reflecting inherently high-dimensional nature and CO2 channeling dataset has sparsity. Therefore, it is necessary to introduce a data-driven transfer learning framework to precisely predict the timing of CO2 channeling. Our proposed framework heavily relies on the Extreme Gradient Boosting (XGBoost) algorithm. Source domains were constructed by learning from various scenarios gas channeling data. The knowledge learned by the source domain model is transferred, allowing for high-precision predictions by simply adjusting parameters for the target domain model. This approach leads to a more comprehensive and accurate analysis of CO2 channeling times. The model was trained on 120 actual reservoir and 200 simulation well datasets tested on the 18 well datasets of target domain, achieving an average R2 value of 0.972 and a MSE value is 2393. Distinguished from numerical simulation and empirical formulas, this work presents a novel, swift, and precise to forecasting CO2 channeling, offering valuable insights for reservoir engineers in managing CO2 channeling prevention and mitigation strategies.

Keywords CO2 channeling, Transfer learning framework, XGBoost algorithm, CO2 enhanced oil recovery