基于深度学习的陆相页岩油储层可压性评价方法
A method for fracability evaluation of continental shale oil reservoir based on deep learning
梅雨 1姜旭 2杨文博1
作者信息
- 1. 西安石油大学石油工程学院,陕西西安 710065;陕西省油气井及储层渗流与岩石力学重点实验室西安石油大学,陕西西安 710065
- 2. 陕西延长石油(集团)有限责任公司研究院,陕西西安 710065
- 折叠
摘要
考虑到传统卷积神经网络(CNN)在提取局部特征方面的优异性能,双向长短期记忆神经网络(BiLSTM)能够更好地捕捉测井数据中的双向依赖关系和地质变化,以及注意力机制增强重要信息的感知能力,将三者的优势结合起来,提出了深度混合神经网络(CBiLSTM-AT).实验结果表明,与传统方法相比,基于均匀流形逼近与投影(UMAP)的可压性评价方法更适合陆相页岩油储层.CBiLSTM-AT优于传统的神经网络模型,实现了更高的预测精度.所提出的方法更高效、更准确,为储层改造提供了有效的技术支持.
Abstract
Considering the excellent performance of traditional convolutional neural network(CNN)in extracting local features,the ability of bidirectional long and short-term memory neural network(BiLSTM)to better capture bidirectional dependencies and geological varia-tions in logging data,as well as the attentional mechanism to enhance the perceptual ability of important information,the advantages of all three are combined,and a deep hybrid neural network(CBiLSTM-AT)is proposed.The experimental results show that the fracability e-valuation method based on uniform manifold approximation and projection(UMAP)is more suitable for continental shale oil reservoirs than the traditional method.CBiLSTM-AT outper-forms the traditional neural network model and achieves higher prediction accuracy.The proposed method is more efficient and accurate,and provides effective technical support for reservoir stimulation.
关键词
陆相/页岩油/可压性/深度学习Key words
continental/shale oil/fracability/deep learning引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据