基于主成分分析的离散过程神经网络水淹层动态预测方法  

Dynamic Recognition Method for Water-flooded Layer with Discrete Process Neural Network Based on the Principal Component Analysis

作　　者：钟仪华[1] 李榕[1] 张志银[1] 朱海双[1] 

机构地区：[1]西南石油大学理学院,四川成都610500

出　　处：《测井技术》2010年第5期432-436,共5页Well Logging Technology

基　　金：四川省教育厅重点项目(07ZA143)资助

摘　　要：提出了一种利用主成分分析和离散过程神经网络进行水淹层动态预测的方法,对测井曲线信息随油层厚度变化的离散数据进行主成分分析,减少了离散过程神经网络模型的输入参数,排除了各参数之间的相关性。引入了反映深度变化累积效应的输入参数——测井参数曲线层段的不同油层厚度。据此建立的识别模型能够反映出随含水率的上升、深度不同时测井曲线的变化规律。实例研究表明,提出的方法与BP神经网络识别方法、支持向量机方法相比较具有更快的运算速度和更高的识别精度,能够体现出高含水期水淹层的动态变化特征。Proposed is a new dynamic recognition method of water-flooded layer with discrete process neural network based on the principal component analysis. The principal component analysis is used in discrete logging information data of various reservoir thickness to remove the correlation of the input samples, and reduce the model input parameters. New dynamic recognition model of water-flooded layer adds reservoir thickness to be a new variable input parameter relating to the depth, which could reflect the variation of logging parameter curve with the rise of moisture content at different depths. The results show that both the operation speed and recogni- tion accuracy of the new dynamic recognition method of water-flooded layer in log interpretations are all better than other recognition methods such as BP neural network identification method and SVM method. This new method could sufficiently reflect the dynamic characteristics of waterflooded layer in high water cut oil field.

关 键 词：测井曲线 动态预测 水淹层识别 主成分分析 离散过程神经网络 

分 类 号：P631.84[天文地球—地质矿产勘探]