摘要
受自然电位测井原理限制,存储式测井不能完成自然电位测井。工程上的传统做法是对GR曲线强滤波,生成自然电位曲线,但这种做法不能反映地层的真实情况。因此,提出一种基于广义回归神经网络的自然电位测井曲线生成方法,通过基于自适应k均值聚类的训练数据清洗和基于最近邻的训练数据自动选择机制,及"成对-合并"策略,综合利用多口邻井的测井数据生成目标井的自然电位曲线。通过在LP油区砂泥岩地层目标井的实验结果对比,相比于传统的BP神经网络,自然电位生成曲线与实际曲线的相关系数提升了0.22,平均相对误差降低3%,证实了本文方法的有效性。本文方法扩展了存储式测井的测量曲线类型及应用范围,使其更好的服务于油田的生产开发。
Wireless memory logging cannot acquire spontaneous potential log data under the limit of spontaneous potential log theory. A SP log curve generation method based on generalized regression neural network(GRNN) is proposed in this paper comprehensively utilizing log data from multiple nearby wells through training data washing mechanism based on k-mean clustering, training data automatically selecting mechanism based on nearest neighbor and "pair-merge" strategy. Experiments in shale-sand target well locating in LP oil block show correlation coefficient is improved by 0.22 and mean relative error is decreased by 3% between generated curve by the method and actual curve compared with traditional BP neural network method. Measured curves category and application of wireless memory logging are extended to provide better services for producing and development of oil fields.
作者
赵自民
Zhao Zimin(North China Branch,China Petroleum Logging CO.LTD.,Renqiu 062552,China)
出处
《电子测量技术》
2020年第7期103-107,共5页
Electronic Measurement Technology
关键词
存储式测井
广义回归神经网络
自然电位
wireless memory well logging
generalized regression neural network
spontaneous potential
