基于多特征联合的BP神经网络岩性识别  

Lithologic Identification of BP Neural Network Based on Multi-feature Combination

作　　者：曹志民[1,2] 王振涛[1] 韩建 李雨昭[1] CAO Zhi-min;WANG Zhen-tao;HAN Jian;LI Yu-zhao(College of Electronic Science, Northeast Petroleum University;Post-doctoral Research Center, Daqing Oilfield)

机构地区：[1]东北石油大学电子科学学院 [2]大庆油田博士后科研工作站

出　　处：《化工自动化及仪表》2018年第5期364-367,395,共5页Control and Instruments in Chemical Industry

基　　金：国家自然科学基金项目(51574087)

摘　　要：为了准确地实现对岩性的有效表征,在原始测井数据基础上,首先对统计特征、纹理特征等对岩性具有更好表征能力的特征进行提取;然后对提取的特征信号进行离散余弦变换(DCT)获取可靠性更高的相关低频特征;最后,利用反向传播神经网络(BP)构建自动岩性识别系统。为了验证所提方法的有效性,采用大庆油田某区块多井测井数据进行了多组交叉对比实验。实验结果表明:利用统计特征和纹理特征的多频组合作为神经网络的输入数据,岩性识别精度明显提高。For purpose of characterizing the lithology accurately, having primitive log data based to extract teatures such as the statistical teature and texture teature which reflecting the lithology better was implemented, including having discrete cosine transtorm（DCT） applied to the characteristic signals extracted to obtain more reliable correlated low-frequency teatures and having backpropagation neural network （BP） adopted to build an automatic lithology identification system. In order to verily effectiveness of the method proposed, the multi- well cross-correlation experiments were conducted by using multi-well log data in Daqing Oilfield. The experi- mental results show that, making use of the multi-frequency combination of statistical teatures and texture tea- tures as the input data of the neural network can obviously improve accuracy of the lithologic identification.

关 键 词：特征提取 反向传播神经网络 测井数据 岩性识别 离散余弦变换 

分 类 号：TN911.72[电子电信—通信与信息系统]