基于双向门控循环单元神经网络的声波测井曲线重构技术  

Acoustic log reconstruction based on bidirectional Gated Recurrent Unit(GRU)neural network

作　　者：周欣 曹俊兴[1,2] 王兴建 王俊[1] 廖万平[1] ZHOU Xin;CAO JunXing;WANG XingJian;WANG Jun;LIAO WanPing(College of Geophysics,Chengdu University of Technology,Chengdu 610059,China;State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation,Chengdu University of Technology,Chengdu 610059,China)

机构地区：[1]成都理工大学地球物理学院,成都610059 [2]油气藏地质及开发工程国家重点实验室(成都理工大学),成都610059

出　　处：《地球物理学进展》2022年第1期357-366,共10页Progress in Geophysics

基　　金：国家自然科学基金重点基金项目“基于地震数据次微成分x-矢量深度学习的四川盆地深层碳酸盐岩储层含气性检测理论方法研究”(42030812);国家自然科学基金“基于深度域地震波频散反演的四川盆地深层碳酸盐岩储层含气量预测理论方法研究”(41974160);国家自然科学基金“基于相干成像的测井远探测高精度成像理论研究联合资助”(42074163)联合资助。

摘　　要：声波测井作为测井与地震资料之间的关键桥梁,对储层岩性、物性分析及定量化评价具有十分重要的意义,完整的声波测井资料有助于获得高分辨率反演剖面,可为储层地质解释提供可靠的依据.然而实际开采过程中很多地区由于仪器故障、井眼垮塌等原因造成声波测井曲线缺失,重新测井不仅价格昂贵而且难以实现.本文拟发展一种基于双向门控循环单元(BGRU)神经网络的声波测井曲线重构技术,对缺失的声波测井曲线进行高效、智能补全.该方法充分考虑了测井序列当前数据与历史和未来数据之间的关联性及测井数据之间的非线性映射关系.将该方法应用于真实测井实验,并将其重构结果与多元回归分析(MLR)和门控循环单元(GRU)神经网络预测结果对比分析,结果表明BGRU神经网络取得了优异的声波测井曲线重构效果,为声波测井曲线预测提供了一条新思路.As a key bridge between well logs and seismic data, acoustic logging is of great significance for reservoir lithology and physical property analysis and quantitative evaluation. Complete acoustic logging data is helpful to obtain high-resolution inversion profiles and provide reliable basis for reservoir geological interpretation. However, due to instrument failure, borehole collapse and other reasons, acoustic logging curves are missing in many areas in the actual mining process, so relogging is expensive and difficult to be realized. This paper intends to develop a acoustic logging curve reconstruction technology based on Bidirectional Gated Recurrent Unit(BGRU) neural network to perform efficient and intelligent completion of the missing acoustic logging curves. This method fully considers the correlation between current log data and historical and future log data and the nonlinear mapping relationship between log data. The method is applied to real logging experiments, and the reconstructed results are compared with the prediction results of Multiple Linear Regression(MLR) and Gated Recurrent Unit(GRU) neural network. The results show that the BGRU neural network achieves excellent acoustic log reconstruction effect, which provides a new idea for acoustic log prediction.

关 键 词：声波测井 重构 非线性映射 循环神经网络 双向门控循环单元神经网络 

分 类 号：P631[天文地球—地质矿产勘探]