石油学报 ›› 2017, Vol. 38 ›› Issue (12): 1425-1433.DOI: 10.7623/syxb201712010

• 油田开发 • 上一篇    下一篇

基于p-stable LSH的多点地质统计建模算法

喻思羽1, 李少华1, 王端平2, 王军2, 张以根2, 于金彪2   

  1. 1. 长江大学地球科学学院 湖北武汉 430100;
    2. 中国石油化工股份有限公司胜利油田分公司勘探开发研究院 山东东营 257015
  • 收稿日期:2017-03-24 修回日期:2017-10-09 出版日期:2017-12-25 发布日期:2018-01-05
  • 通讯作者: 李少华,男,1972年8月生,1994年获江汉石油学院学士学位,2003年获中国石油勘探开发研究院博士学位,现为长江大学地球科学学院教授、博士生导师,主要从事地质统计学、地质建模方面的研究和教学。Email:534354156@qq.com
  • 作者简介:喻思羽,男,1987年6月生,2009年获长江大学学士学位,现为长江大学博士研究生,主要从事地质统计学算法的研究。Email:573315294@qq.com
  • 基金资助:

    国家自然科学基金项目(No.41572121)、国家重大科技专项(2016ZX05011-001)和湖北省科技创新群体项目(2016CFA024)资助。

Multipoint geo-statistical modeling algorithm based on p-stable LSH

Yu Siyu1, Li Shaohua1, Wang Duanping2, Wang Jun2, Zhang Yigen2, Yu Jinbiao2   

  1. 1. College of Geosciences, Yangtze University, Hubei Wuhan 430100, China;
    2 Research Institute of Exploration and Development, Sinopec Shengli Oilfield Company, Shandong Dongying 257015, China
  • Received:2017-03-24 Revised:2017-10-09 Online:2017-12-25 Published:2018-01-05

摘要:

SIMPAT将图像重建思想引进储层地质建模领域,借助于弱化概率的相似性判别指标,用最相似地质模式替换待估点处的数据事件完成预测。当模型较大且数据样式较多时,海量的数据样式相似度计算使得SIMPAT的计算效率较低。为了有效平衡多点地质统计建模算法效率和内存的矛盾,基于SIMPAT算法,提出基于p-stable局部敏感哈希的多点地质统计建模算法LSHSIM,该方法使用局部敏感哈希将数据样式的特征向量映射到哈希表。建模时从哈希表里取出与数据事件的特征向量具有相同哈希值的数据样式,用最相似的数据样式替换覆盖待估区的数据事件完成建模。利用实例对比新算法与SIMPAT等现有方法的结果表明,LSHSIM算法计算效率高,并节省了内存空间,对算法的关键参数进行了敏感性分析、非条件和条件模拟,能较好再现训练图像的先验地质模式。

关键词: 储层建模, 局部敏感哈希, SIMPAT, 多点地质统计学, 训练图像

Abstract:

Image reconstruction idea is introduced into reservoir geological modeling field by SIMPAT, of which the most similar geological mode is used to replace the data event at a to-be-estimated point for prediction using the similarity evaluation index of weakening probability. When the model is larger with more data patterns, the SIMPAT computational efficiency is lower due to the humongous data pattern similarity calculation. To effectively balance the contradiction between the efficiency and memory of multipoint geo-statistical modeling algorithm, based on the SIMPAT algorithm, the multipoint geo-statistical modeling algorithm LSHSIM based on p-stable locality sensitive hashing was put forward, and in this method, the locality sensitive hashing was used for mapping the eigenvector of data pattern on the hash table. During modeling, the data pattern with the same hash value to the eigenvector of data event was extracted from the hash table, and the most similar data pattern was applied to replace the data events at the to-be-estimated area. Through the instance-based comparison between the new algorithm and existing methods such as SIMPAT, and the new algorithm has a high computational efficiency in saving the memory space, and the sensibility analysis, non-conditional and conditional simulations are conducted on the key parameters of algorithm, which can better reproduce the prior geo-mode of training image.

Key words: reservoir modeling, local sensitive hashing, SIMPAT, multipoint geo-statistic, training image

中图分类号: