Abstract: In the study area, thin interbeds of sparite and mudstone are mainly developed in the lower submember of Member 3 and upper submember of Member 4 of Shahejie Formation in the Niuzhuang subsag, which are effective enrichment areas and stable production channels for shale oil. However, the insufficient resolution of conventional series of logging leads to great difficult in identifying thin interbeds. To address this issue, a hybrid model of extreme learning machine with particle swarm optimization is used to improve the accuracy of identification for thin interbeds. A PSO-ELM-based identification model for thin interbeds is constructed by selecting 8 conventional logging parameters and 3 high-resolution logging curves that reflect the "three qualities" of reservoirs as physical constraints. The results show that compared with the common machine learning models such as ELM, SVM, and BP, the proposed PSO-ELM machine learning model is more stable, of which the identification accuracy of thin interbeds is improved by 10% to 30%, and can more precisely describe the thin interbeds with a thickness of about 0.3 m, providing technical support for further shale oil exploration and development.

Key words: shale oil, thin interbeds, particle swarm optimization, extreme learning machine, Niuzhuang subsag

摘要： 东营凹陷牛庄洼陷沙河街组三段下亚段和沙河街组四段上亚段主要发育重结晶灰岩、泥岩薄互层，是页岩油的有效富集区和稳定产出通道，但由于常规测井系列分辨率不足，导致薄互层识别难度大。针对这一问题，采用粒子群（PSO）优化的极限学习机（ELM）混合模型以提升薄互层识别准确率，选取反映储层"三品质"特征的8条常规测井参数及3条高分辨率测井曲线作为物理约束，构建了基于PSO-ELM的薄互层识别模型。研究结果表明，与ELM、支持向量机、前馈神经网络等常见的机器学习模型相比，所提出的PSO-ELM机器学习模型稳定性更强，薄互层识别准确率提升幅度为10%~30%，且更能精准刻画厚度约为0.3 m的薄互层，该方法可以为页岩油勘探开发提供一定技术支持。

关键词: 页岩油, 薄互层, 粒子群优化, 极限学习机, 牛庄洼陷