基于模式识别融合的低阻油层识别

Low-resistance Oil Layer Recognition Based on Pattern Recognition Fusion

在油田开发生产过程中会产生大量测井、录井以及解释生产数据,且产生的海量数据因时间、地点以及记录方式的不同,其数据表现形式也千差万别。测井数据作为石油勘探开发研究的基石,在油藏识别预测方面发挥着重要作用。由于测井数据受沉积微相、构造、钻井以及上下层等诸多因素作用,其特征表示较为复杂,且不同厚度小层其产生的测井数据样本点也不同,因此如何准确表征测井数据特征并与小层数据融合、实现低阻油层精准识别是一项值得研究的内容。该文结合费雷歇相似性度量方法对小层各测井曲线数据进行特征识别与描述,基于随机森林与XGBoost集成学习方法,以模式识别特征与录井、解释特征为输入,构建低阻油层识别模型,精准识别低阻油层。应用该方法,对港*油田真实数据进行了实例分析,结果表明,基于模式识别融合的低阻油层识别方法能够实现低阻油层的挖潜,识别精准率高达90%,节省人工分析带来的高成本,降低了人工分析的主观性和片面性,该模型的提出是大数据挖掘技术在低阻油层识别的实际应用。

In the process of oilfield development and production,a large number of logging,logging and interpretation produc⁃tion data will be generated.Due to the different time,place and recording method,the data forms are also different.As the corner⁃stone of petroleum exploration and development,logging data plays an important role in reservoir identification and prediction.Due to the influence of sedimentary microfacies,structure,drilling,upper and lower layers and many other factors,the characteristics of logging data are more complex,and the sample points of logging data produced by different thickness layers are also different.Therefore,how to accurately characterize the characteristics of logging data and integrate with small layer data to realize the accu⁃rate identification of low resistivity reservoirs is a content worthy of study.In this paper,combining with the Fisher similarity mea⁃surement method to identify and describe the characteristics of each logging curve data in the small layer,with the pattern of recogni⁃tion features and the pattern of Logging data as modle input,an integrated learning model based on E-Learning of the Random forest and the XGBoost is built to achieve accurate identification of low resistivity reservoir.Using the method in this article,a case analy⁃sis on the real data of the Gang*Oilfield is conducted.The results show that the low-resistance oil layer identification method based on pattern recognition fusion can realize the potential of low-resistance oil layer,the recognition accuracy rate is as high as 90%,the high cost caused by manual analysis is saved,and the subjectivity and one-sidedness of manual analysis are greatly reduced.The proposed model is the practical application of big data mining technology in the identification of low-resistance oil layers.