石油地球物理勘探2025，Vol.60Issue(3)：587-597,11.DOI:10.13810/j.cnki.issn.1000-7210.20240255
基于代价敏感学习的碳酸盐岩储层流体识别
Fluid identification of carbonate reservoirs based on cost-sensitive learning
摘要
Abstract
Fluid identification in carbonate reservoirs is crucial for reservoir assessment and hydrocarbon devel-opment.However,carbonate reservoirs are strongly non-homogeneous,which makes it difficult to realize their accurate identification by traditional methods.Although machine learning-based methods can deeply explore the intrinsic connection between logging data and oil,gas,and water information to improve the identification ef-fect,they are easily affected by the noise of logging data,and the sample category ratio is imbalanced.In this paper,a reservoir fluid identification method based on cost-sensitive learning is proposed with carbonate reser-voirs in the Sichuan Basin as the research object.First,the wavelet transform is used to reduce the noise of log-ging data to solve the data noise problem.Then,the correlation test of logging curves is carried out by integra-ting analysis of variance,decision tree,and reservoir theory to screen out the logging curves that are highly cor-related with the reservoir fluid types.Finally,the neural network model is designed to address sample category imbalance by using the cost-sensitive learning strategy,so as to improve identification accuracy.The results show that the wavelet transform reduces data noise and improves the accuracy of fluid identification.The log-ging curves AC,CNL,CAL,RT,GR,and RXO are highly correlated with the fluid types in carbonate reser-voirs.The cost-sensitive learning method effectively addresses the problem of low identification accuracy of a few classes due to the imbalanced data,and the identification accuracy of the model reaches 97.61%,which is better than that of other comparative models.It provides a feasible solution for fluid identification in carbonate reservoirs.关键词
储层流体/代价敏感/碳酸盐岩/小波变换/机器学习/测井参数Key words
reservoir fluids/cost-sensitive/carbonate rocks/wavelet transform/machine learning/logging para-meters分类
地质学引用本文复制引用
高国海,赵祥东,蒋薇,王杨,刘勇,王欣..基于代价敏感学习的碳酸盐岩储层流体识别[J].石油地球物理勘探,2025,60(3):587-597,11.基金项目
本项研究受国家自然科学基金优秀青年基金项目"页岩气多尺度非线性渗流力学"(52222402)、油气藏地质及开发工程国家重点实验室开放性研究课题"面向油气产量预测的非平稳时间序列模型构建及应用"(PLN2022-33)和南充市科技计划项目"储层自动识别及分类评价技术研究"(23XNSYSX0111)联合资助. （52222402）
