摘要
地层识别是油气藏勘探的研究基础。传统地层识别由地质学家根据自身掌握的知识和经验手工完成,这种地质学家主导的人工解释是主观的、耗时的,可能引入人为偏差。深度学习在解决复杂非线性问题上具有优势,目前尚无有效解决地层识别的深度学习方法。针对测井-地层识别,提出了基于特征工程和一维卷积神经网络的地层智能识别方法。首先,利用INPEFA技术和中值滤波对原始曲线进行了多维重构,更好地提取了原始曲线的地层趋势及边缘特征,并对重构矩阵和原始曲线特征采用K-means聚类算法提取时空相关聚类特征;然后,以原始曲线特征、INPEFA曲线、中值滤波特征和聚类特征作为输入,基于一维卷积神经网络得到当前深度地层预测类型。与长短期记忆网络(LSTM)和传统的机器学习方法对比发现,在地层的识别上,地层智能识别方法具有更优异的性能和鲁棒性。该方法能有效识别地层,识别准确率达到92.82%,且在识别地层的同时也完成了地层划分。
Stratigraphic recognition is the basis for the research of oil and gas reservoir exploration.Traditional stratigraphic identification is done manually by geologists based on their own knowledge and experience,and this geologists-led manual interpretation is subjective,time-consuming,and can introduce artificial bias.Deep learning has advantages in solving complex nonlinear problems,and there is currently no effective deep learning method to solve formation recognition.For logging-stratigraphic recognition,a stratigraphic intelligent recognition method based on feature engineering and one-dimensional convolutional neural network is proposed.Firstly,the original curve is reconstructed by INPEFA and median filtering,the stratigraphic trend and edge features of the original curve are better extracted,and the K-means clustering algorithm is used to extract the spatiotemporal correlation clustering features of the reconstructed matrix and the original curve.Then,taking the original curve features,INPEFA curves,median filtering features and clustering features as inputs,the current deep stratigraphic prediction type is obtained based on the one-dimensional convolutional neural network.Compared with the long short-term memory network(LSTM)and traditional machine learning methods,the formation intelligent recognition method has better performance and robustness in the recognition of strata.The proposed method can effectively identify strata,the recognition accuracy reaches 92.82%,and the strata division is completed at the same time as identifying the strata.
作者
曹茂俊
崔欣锋
CAO Mao-jun;CUI Xin-feng(School of Computer and Information Technology,Northeast Petroleum University,Daqing 163318,China)
出处
《计算机技术与发展》
2023年第9期133-140,148,共9页
Computer Technology and Development
基金
黑龙江省自然科学基金(LH2019F004)
东北石油大学优秀中青年科研创新团队(KYCXTD201903)。