基于集成学习的油藏井筒一体化智能诊断模型

Oil reservoir-well-integrated intelligent diagnostic model based on ensemble learning

杨耀忠 邴绍强 马承杰 于金彪 王相 李秉超 景瑞林 孙召龙

YANG Yaozhong BING Shaoqiang MA Chengjie YU Jinbiao WANG Xiang LI Bingchao JING Ruilin SUN Zhaolong

中国石化胜利油田分公司 常州大学石油与天然气工程学院

SINOPEC Shengli Oilfield Company, Dongying 257000, Shandong, China School of Petroleum and Natural Gas Engineering, Changzhou University, Changzhou 213164, Jiangsu, China

目前油藏、采油依托各自专业数据和信息系统进行异常问题的分析，对于两个系统间的复杂关联关系考虑不够，导致生产异常的诊断仍较局限，治理措施针对性不强。基于随机森林算法和卷积神经网络算法集成学习构造了油藏井筒一体化智能诊断模型，根据注水失效、泵漏失等不同油藏、井筒问题，以基于随机森林的决策树分析油藏异常工况，卷积神经网络诊断井筒异常故障，通过集成学习方法将两类分类器结合起来，形成一体化诊断。现场验证结果表明，所建立的方法通过集成学习提升了单分类器性能与范化能力，应用准确率达到90%以上，实现了油藏和井筒问题的一体化诊断，为油田智能化管控提供了有力支撑。

At present,reservoir engineering and production engineering analyze anomalies separately, using specialized data and information systems of their disciplines. In such cases, the complex correlation between the reservoir and well systems is somewhat simplified and thus the resultant diagnosis of production anomalies is inadequate and the proposed treatment lacks pertinence. Given this, the oil reservoir-well-integrated intelligent diagnostic model, based on the ensemble learning consisting of the random forest and convolution neural network (CNN), was developed. For different oil reservoir and well anomalies such as ineffective water injection and pump leakage, the decision-making tree based on the random forest was used to investigate abnormal conditions of oil reservoirs, the CNN was adopted to diagnose well anomalies, and finally, the integrated diagnosis was produced by integrating the two classifiers via ensemble learning. The field validation of the diagnosis showed the presented method improves the performance and normalization capacity of the constituent classifier via ensemble learning and the integrated diagnosis of reservoir and well anomalies is delivered, with an application accuracy of over 90%. This invention can strongly support the intelligent management of oilfields.

10.13639/j.odpt.2022.03.017