摘要：

细粒沉积岩岩相的空间展布特征是页岩油勘探的关键研究内容。由于高成本且稀缺的取心井限制了岩相的分析,测井预测岩相的工作变得尤为重要。以松辽盆地青山口组一段(青一段)为例,建立了“岩性+矿物+TOC+沉积构造”的岩相划分方案,确定了7类岩相。结合6条常规测井曲线形成了X8井的岩相—测井数据库。综合使用机器学习中的随机森林(RF)、XGBoost和支持向量机(SVM)3种模型评价岩相的预测效果,并确定SVM是最优分类模型,使用BSMOTE处理岩相样本的分类不平衡问题并将处理后的数据输入到SVM模型,建立了BSMOTE-SVM的岩相预测的组合模型。BSMOTE-SVM的预测效果最佳,准确率(A)、精确率(P)、召回率(R)和F1依次分别为86.49%、 86.60%、 86.49%和86.31%。该组合模型可快速且精准的预测多井的岩相,并确定了松辽盆地长岭凹陷青一段的岩相分布,为下一步页岩油有利富集区的优选提供了一定指导依据。

关键词: 细粒沉积岩, 岩相, 测井预测, BSMOTE-SVM, 青山口组, 松辽盆地

Abstract:

The spatial distribution of lithofacies of fine-grained sedimentary rocks is a critical research focus in shale oil exploration. Due to the high cost and scarcity of core wells,which constrain direct lithofacies analysis,logging-based prediction has become increasingly essential. Taking the First Member of the Qingshankou Formation in the Songliao Basin as a case study,this research establishes a lithofacies classification scheme integrating lithology,mineral composition,total organic carbon(TOC),and sedimentary structures,resulting in the identification of seven distinct lithofacies types. A lithofacies-well log dataset was constructed for Well X8 using six conventional logging curves. Three machine learning algorithms—Random Forest(RF),eXtreme Gradient Boosting(XGBoost),and Support Vector Machine(SVM)—were employed to evaluate classification performance,with SVM identified as the optimal model. To address class imbalance in the training data,the BSMOTE(Borderline Synthetic Minority Oversampling Technique)algorithm was applied. The balanced dataset was then used to develop a hybrid lithofacies prediction model: BSMOTE-SVM. The BSMOTE-SVM model demonstrated the best predictive performance,achieving an accuracy of 86.49%,precision of 86.60%,recall of 86.49%,and F1-score of 86.31%. This integrated model enables rapid and accurate lithofacies prediction across multiple wells and delineates the lithofacies distribution in Member 1 of the Qingshankou Formation in the Changling sag,offering a robust foundation for selecting favorable shale oil enrichment zones in future exploration.

Key words: fine-grained sedimentary rocks, lithofacies, logging prediction, BSMOTE-SVM, Qingshankou Formation, Songliao Basin

中图分类号: 

• P618.130