Seddiqi Khwaja Naweed, Hao Hongda, Liu Huaizhu, Hou Jirui
The Unconventional Oil and Gas Institute, China University of Petroleum-Beijing, Changping, Beijing 102249, China.
School of Petroleum Engineering, Changzhou University, 21 Gehu Zhong Road, Changzhou, Jiangsu 213164, China.
ACS Omega. 2021 Dec 6;6(50):34327-34338. doi: 10.1021/acsomega.1c03973. eCollection 2021 Dec 21.
The major oil fields are currently in the middle and late stages of waterflooding. The water channels between the wells are serious, and the injected water does little effect. The importance of profile control and water blocking has been identified. In this paper, the decision-making technique for water shutoff is investigated by the fuzzy evaluation method, FEM, which is improved using a random forest, RF, classification model. A machine learning random forest algorithm was developed to identify candidate wells and to predict the well performance for water shutoff operation. A data set consisting of 21 production wells with three-year production history is used, where out of the mentioned well data, 70% of them are implemented for training and the remaining are used for testing the model. After fitting the model, the new weights for the factors are established and decision-making is made. Accordingly, 16 wells out of 21 wells are selected by the FEM where 8 wells out of 21 wells are selected by the new factor weight created by RF for water shutoff. A numerical simulation model is established to plug the selected wells by both methods after which the influence of plugging on water cut, daily oil production, and cumulative oil production is compared. The paper shows that the reservoir had a better performance after eight wells were selected using a new weighting system created by RF instead of the 16 wells that were selected using the FEM model. The paper also states that the new weighting model's accuracy improved the decision-making abilities of the wells.
主要油田目前处于注水开发的中后期。井间水道严重，注水效果不佳。已认识到调剖堵水的重要性。本文采用模糊评价法（FEM）研究堵水决策技术，并利用随机森林（RF）分类模型对其进行改进。开发了一种机器学习随机森林算法，用于识别候选井并预测堵水作业的油井性能。使用了一个由21口具有三年生产历史的生产井组成的数据集，在上述井数据中，70%用于训练，其余用于测试模型。拟合模型后，建立新的因素权重并进行决策。因此，通过FEM从21口井中选出了16口井，通过RF创建的新因素权重从21口井中选出了8口井进行堵水。建立了数值模拟模型，用两种方法对选定的井进行封堵，然后比较封堵对含水率、日产油量和累计产油量的影响。本文表明，使用RF创建的新权重系统选择8口井后，油藏性能优于使用FEM模型选择的16口井。本文还指出，新的权重模型的准确性提高了油井的决策能力。
