鄂尔多斯盆地延长组多油源贡献比例与分布规律:基于机器学习与可解释性研究  

Contribution ratio and distribution patterns of multiple oil sources in the Yanchang Formation of the Ordos Basin:A study utilizing machine learning and interpretability techniques

作　　者：苏恺明 徐耀辉 徐旺林[3] 张月巧[3] 白斌[3] 李阳 严刚 SU Kaiming;XU Yaohui;XU Wanglin;ZHANG Yueqiao;BAI Bin;LI Yang;YAN Gang(Hubei Key Laboratory of Petroleum Geochemistry and Environment,Wuhan 430100,China;College of Resources and Environment,Yangtze University,Wuhan 430100,China;Research Institute of Petroleum Exploration and Development,PetroChina,Beijing 100083,China)

机构地区：[1]油气地球化学与环境湖北省重点实验室,湖北武汉430100 [2]长江大学资源与环境学院,湖北武汉430100 [3]中国石油勘探开发研究院,北京100083

出　　处：《地学前缘》2024年第3期530-540,共11页Earth Science Frontiers

基　　金：中国博士后科学基金面上资助(2023M730365);湖北省自然科学基金计划青年项目(2023AFB232);中石油科学研究与技术开发项目(2021DJ0404)。

摘　　要：鄂尔多斯盆地延长组发育多套潜在的烃源岩,但不同烃源岩之间生物标志物特征相似,常规油源对比方法效果不佳,相关认识长期存在争议。基于这样的问题,本文提出了一种基于深度学习的油源对比方案,将人工智能方法应用于油源对比研究,所开展的工作和认识有:(1)以延长组不同层位大量泥岩、页岩样品的42种生物标志物参数作为学习数据,构建了一种识别未知样品油源类别的深度神经网络模型,对长7泥页岩、长8—长10泥页岩的判别正确率分别达到了79.6%和83.0%,实现了延长组主要烃源岩生烃产物的有效区分;(2)通过模型分析了大量砂岩、原油样品的油源分类,统计了不同烃源岩对于延长组各个油层组原油的贡献比例,总结了它们的分布规律;(3)基于目前较为先进的置换特征重要性(PFI)算法,对所得模型进行了敏感性分析,初步揭示了延长组两类主要烃源岩的生物标志物差异。本文对于人工智能方法、技术在石油分子地球化学领域的发展具有积极的参考价值。The Yanchang Formation within the Ordos Basin hosts multiple sets of potential source rocks,all exhibiting similar biomarker properties.The conventional method of oil-source correlation has proven ineffective,leading to longstanding debates within the field.In response to these challenges,this study introduces a novel deep learning-based scheme for oil-source comparison,leveraging artificial intelligence methods for research in this domain.The study presents the following key findings and insights:(1)Development of a deep neural network model for identifying the oil source type of unknown samples by utilizing 42 biomarker parameters from a diverse set of mudstone and shale samples representing different oil groups within the Yanchang Formation as training data.The model achieved identification accuracies of 83.0%for Chang 7 mudstone and 79.6%for Chang 8 Chang 10 mudstone,successfully distinguishing the primary source rocks of the Yanchang Formation from hydrocarbon generation products.(2)Application of the model to analyze the oil source classification of numerous sandstone and oil samples.The study calculated the contribution ratios of various source rocks to each oil group within the Yanchang Formation,summarizing their distribution patterns.(3)Conducting sensitivity analysis of the model using the permutation feature importance(PFI)algorithm,revealing differences in biomarkers between the two main source rocks of the Yanchang Formation.These findings contribute to advancing artificial intelligence techniques and technologies in the field of petroleum molecular geochemistry,offering valuable insights for future research and applications.

关 键 词：机器学习 深度神经网络 敏感性分析 鄂尔多斯盆地延长组 油源对比 

分 类 号：TE122.11[石油与天然气工程—油气勘探]