摘要
本文旨在开发一种新的基于集成学习的火成岩储集层流体识别方法,以解决全球油气资源开发中火成岩储层因岩性和岩相多变、储集空间复杂等因素导致的传统流体识别方法效果不佳的问题。该方法创新性地将自适应多目标群体优化算法(AMSCO)与基于极端梯度提升(XGBoost)的深度森林算法相结合,利用常规测井数据集,对复杂岩性火成岩储层进行流体识别。在方法设计上,本研究首先采用AMSCO算法对不平衡常规测井数据集进行优化,有效解决了数据集中存在的类别不平衡问题,为后续模型训练提供了更为均衡的数据基础。其次,充分利用XGBoost在处理大规模数据和高维度数据上的优势,以及深度森林在特征提取和分类任务上的卓越性能,构建了一个融合了XGBoost和深度森林的高效的集成学习模型CXDF(cross-adaptive XGBoost and deep forest),从而实现了对复杂岩性火成岩储层流体的准确识别。为验证该方法的有效性,本文将其与支持向量机(SVM)、XGboost和基于XGboost的深度森林一起应用于模拟井中进行模型验证比较,并将模型应用于实际地层。结果表明,该方法在模拟井中的各项评价指标均优于其他方法,特别是在识别非产水层流体时表现出更高的准确性。在实际地层的应用中,该方法在不同流体结构的储层中均保持了较高的识别性能,展现出良好的泛化能力和稳定性。
A novel integrated learning-based method for igneous reservoir fluid identification is proposed to address the limitations of traditional approaches in handling complex lithological variations and heterogeneous reservoir spaces,which are crucial for global oil and gas resource development.In this paper,the adaptive multi-objective swarm crossover optimization(AMSCO)innovatively combined with an engineered extreme gradient boosting(XGBoost)based on deep forest method for fluid identification in complex lithologic igneous reservoirs using conventional logging data set.Methodologically,firstly,the AMSCO algorithm is used to optimize the imbalanced conventional logging data set,effectively solving the problem of class imbalance in the data set,providing a more balanced data basis for subsequent model training.Secondly,a cross-adaptive XGBoost and deep forest(CXDF)is constructed by fully utilizing XGBoost's advantages in processing large-scale and high-dimensional data,as well as the excellent performance of deep forest in feature extraction and classification tasks.Thus,the accurate identification of reservoir fluids in complex lithologic igneous rocks is achieved.Then,to verify the effectiveness of this method,the model was applied to the simulated well together with support vector machine(SVM),XGBoost and XGBoost-based deep forest for comparison.Finally,the model is applied to the actual stratum.The results show that the evaluation index of the proposed method in the simulated well is superior to other methods,especially in the identification of non-water-producing reservoir fluids.In the application to actual formations,this method maintains high identification performance in different reservoirs with different fluid structures,and shows good generalization ability and stability.
作者
陆哲昆
王祝文
韩锐羿
王欣如
李岩
张鹏济
于宏达
崔裔曈
LU Zhekun;WANG Zhuwen;HAN Ruiyi;WANG Xinru;LI Yan;ZHANG Pengji;YU Hongda;CUI Yitong(Jiangxi Traffic Design and Research Institute Co.,Ltd.,Nanchang 330002,China;College of Geo-Exploration Science and Technology,Jilin University,Changchun 130026,China;School of Geomatics and Prospecting Engineering,Jilin Jianzhu University,Changchun 130046,China)
出处
《世界地质》
2025年第3期481-494,共14页
World Geology
基金
国家自然科学基金项目(41874135)。
关键词
机器学习
集成学习
储层流体识别
火成岩
machine learning
ensemble learning
reservoir fluid identification
igneous rock
