BP神经网络在岩相预测中的应用  

Application of BP neural network in seismic-lithofacies prediction

作　　者：周传友 李少华[1] 何维领 丁芳 黄鑫 ZHOU Chuanyou;LI Shaohua;HE Weiling;DING Fang;HUANG Xin(School of Geosciences, Yangtze University, Wuhan 430100, China;Shanghai Branch of CNOOC Ltd., Shanghai 200030, China)

机构地区：[1]长江大学地球科学学院,武汉430100 [2]中海油上海分公司,上海200030

出　　处：《物探化探计算技术》2020年第6期743-751,共9页Computing Techniques For Geophysical and Geochemical Exploration

基　　金：国家科技重大专项(2016ZX05027-004);国家自然科学基金(41872129)。

摘　　要：N气田花港组上段H3气层组为主要的“甜点”分布区,H3b砂层组最为显著,属于低渗气层。该区域储层非均质性强,目前仅有四口探井。为了建立相对可靠的岩相模型,需要利用地震属性约束井间岩相的预测。研究区单一地震属性与井上岩相的相关性均较差,采用BP神经网络方法,利用多种地震属性组合,提高了单井岩相预测的准确性。对BP神经网络中的隐含层数、训练集和验证集样本数进行了敏感性分析,确定了适合研究区的计算方案,为三维岩相模型的建立提供了借鉴。The H3 upper Huagang gas formation of N gas field is the main"sweet spot"distribution area.H3b sand formation is the most remarkable one which belongs to low permeability gas reservoir.The heterogeneity of the reservoir is high.Up to now,only four wells have been drilled.In order to establish a relatively reliable lithofacies model,the prediction of interwell lithofacies needs to be constrained by seismic attributes.The correlation between single seismic attributes and lithofacies in the study area is poor,thus the paper adopts BP neural network method to address this issue.The result improves the prediction accuracy of single well lithofacies by using multiple seismic attribute assemblages.Sensitivity analysis of hidden layers,training and validation samples in BP neural network is carried out.And the suitable calculation scheme for the study area has been determined.It lays a foundation for the establishment of three-dimensional lithofacies model.

关 键 词：岩相预测 模式识别 BP神经网络 地震属性 H3b砂层组 

分 类 号：P631.4[天文地球—地质矿产勘探]