摘要：

随着页岩气勘探开发的不断深入,精准、高效地划分与预测岩相成为评价页岩气储集层品质的关键。传统的岩相测井识别方法存在主观性强、难以应对大规模数据集以及难以有效捕捉数据中复杂的非线性关系等局限性。本研究以四川盆地下寒武统筇竹寺组页岩气储集层为例,充分利用岩心测试分析的精确数据刻度测井曲线建立的岩相标签数据集,提出了一种融合传统机器学习模型: 决策树、支持向量机(SVM)、K近邻(KNN)以及深度学习模型: 一维卷积神经网络(1D-CNN)的集成学习(Voting)模型进行岩相分类预测,且对比传统机器/深度学习模型与集成学习模型的预测效果,评估其在复杂页岩岩相预测中的应用潜力。研究结果表明: 集成学习(Voting)模型在岩相预测精度上表现最佳,预测精度达到94.0%,显著优于单一基学习器模型(决策树、KNN、SVM及1D-CNN)。形成的集成学习方法,通过融合4种基学习器模型各自的优势,有效克服了传统分类算法受岩相样本不均衡的影响引起的识别偏差,提升了稀缺样本的岩相识别能力,为四川盆地高—过成熟筇竹寺组深层页岩气储集层测井岩相剖面的高精度恢复及甜点预测提供了技术支撑。

关键词: 筇竹寺组, 岩相预测, 机器学习, 深度学习, 集成学习

Abstract:

As the exploration and development of shale gas continue to deepen,the accurate and efficient division and prediction of lithofacies have become key factors in evaluating the quality of shale gas reservoirs. Traditional lithofacies logging identification methods are limited by their strong subjectivity,inability to handle large-scale datasets,and difficulty in effectively capturing the complex nonlinear relationships in the data. This study takes the Lower Cambrian Qiongzhusi Formation shale gas reservoir in the Sichuan Basin as an example. It utilizes precise data from rock core testing and logging curves to establish a lithofacies-labeled dataset. A novel ensemble learning(Voting)model is proposed,integrating traditional machine learning models(Decision Tree,Support Vector Machine(SVM),and K-Nearest Neighbors(KNN))and a deep learning model(1D Convolutional Neural Network,1D-CNN)for lithofacies classification and prediction. The study also compares the predictive performance of traditional machine learning and deep learning models with that of the ensemble learning model,assessing its application potential in complex shale lithofacies prediction. The results indicate that the ensemble learning(Voting)model achieves the highest lithofacies prediction accuracy of 94.0%,significantly outperforming individual base learners(Decision Tree,KNN,SVM,and 1D-CNN). The proposed ensemble learning method effectively overcomes the recognition bias caused by sample imbalance in traditional classification algorithms by leveraging the strengths of the four base learner models. This improvement enhances the ability to identify lithofacies with limited samples,providing technical support for the high-precision reconstruction of logging-based lithofacies profiles and sweet spot prediction in the deeply buried,over-mature Qiongzhusi Formation shale gas reservoirs in the Sichuan Basin.

Key words: Qiongzhusi Formation, lithofacies prediction, machine learning, deep learning, ensemble learning

中图分类号: 

• P618.130.2