欢迎访问《高校地质学报》官方网站,今天是
分享到:

高校地质学报 ›› 2021, Vol. 27 ›› Issue (5): 577-586.DOI: 10.16108/j.issn1006-7493.2020090

• 渤海海域中深层油气精细勘探开发理论技术专栏 • 上一篇    下一篇

莱州湾凹陷垦利油田沙河街组储层渗透率评价模型研究

汪瑞宏1*,齐 奕1,李志愿1,马 超1,王玥天2,3,蔡文浙2,3   

  1. 1. 中海石油(中国)有限公司 天津分公司,天津 300450;
    2. 中国石油大学(北京)油气资源与探测国家重点实验室,北京 102249;
    3. 中国石油大学(北京)地球探测与信息技术北京市重点实验室,北京 102249
  • 出版日期:2021-10-20 发布日期:2021-10-27

Permeability-estimation Model of the Shahejie Formation Reservoir in the KL Oilfield, Laizhou Bay Sag

WANG Ruihong1*,QI Yi1, LI Zhiyuan1,MA Chao1,WANG Yuetian2,3,CAI Wenzhe2,3   

  1. 1. Tianjin Branch of CNOOC Ltd., Tianjin 300459, China;
    2. State Key Laboratory of Oil and Gas Resources and Exploration, China University of Petroleum(Beijing) , Beijing 102249, China;
    3. Beijing Key Laboratory of Earth Exploration and Information Technology, China University of Petroleum(Beijing), Beijing 102249, China
  • Online:2021-10-20 Published:2021-10-27

摘要: 文章以莱州湾凹陷垦利油田沙河街组储层为例,对传统的回归统计模型和基于BP神经网络的人工智能预测模型评价储层渗透率方法和效果进行了对比研究。目标储量报告里定火沙三段中孔、中渗;岩性(粒度)和孔隙度是储层渗透率的主要影响因素。根据岩心及测井数据,建立了孔隙度——粒度二元回归渗透率统计评价模型和BP神经网络渗透率预测模型。通过检验样本集精度对比,分析了隐含层数、隐含层节点数等网络结构参数变化对模型预测结果的影响,重点分析了不同的测井参数输入对BP神经网络模型预测结果的影响。优化后的BP神经网络模型对检验样本集的渗透率预测结果精度最高,其平均相对误差为37%,比传统的二元回归统计模型精度提高了26%。对目标油田三口井连续处理,BP神经网络模型渗透率预测结果更加合理,可以满足开发层段产能分析等生产需求。

关键词: 测井, 渗透率, 统计模型, BP神经网络, 模型预测, 模型优化

Abstract: Based on a case study of the Paleogene Shahejie Formation of the KL oilfield in the Laizhou Bay Sag, this study compares the traditional statistical model of regression with the prediction model of artificial intelligence based on BP neural network to evaluate the reservoir permeability. The target reservoir is mainly of medium to high porosity and permeability. Lithology and porosity are the main influencing factors of reservoir permeability. Based on core data analysis, the porosityparticle size binary regression model and BP neural network were established. By comparing the accuracy of the test sample set, the influence of the network structure parameters such as the number of hidden layers and the number of hidden layer nodes on the prediction results of the BP neural network model was analyzed. The effects of different logging parameters on the prediction results of BP neural network model were analyzed with the focus. The optimized BP neural network model has the highest permeability prediction accuracy for the test sample set, with an average relative error of 37%, which is 26% higher than the traditional statistical model of binary regression. For the continuous treatment of three wells in the target field, the permeability prediction results of the BP neural network model are more reasonable, which can meet the production requirements such as the production capacity analysis of the development zone.

Key words: well log, permeability, statistic model, BP neural network, prediction model, model optimization

中图分类号: