石油学报 ›› 2018, Vol. 39 ›› Issue (8): 916-923.DOI: 10.7623/syxb201808007

• 油田开发 • 上一篇    下一篇

基于图型矢量距离的多点地质统计相建模算法

王鸣川, 段太忠   

  1. 中国石油化工股份有限公司石油勘探开发研究院 北京 100083
  • 收稿日期:2017-10-23 修回日期:2018-05-23 出版日期:2018-08-25 发布日期:2018-09-04
  • 通讯作者: 段太忠,男,1961年8月生,1982年获江汉石油学院学士学位,1996年获挪威科技大学博士学位,现为中国石油化工股份有限公司石油勘探开发研究院国家"千人计划"特聘专家,主要从事储层表征与建模工作。Email:duantz.syky@sinopec.com
  • 作者简介:王鸣川,男,1985年8月生,2007年获中国地质大学(武汉)学士学位,2014年获北京科技大学博士学位,现为中国石油化工股份有限公司石油勘探开发研究院工程师,主要从事储层建模与油藏数值模拟工作。Email:wangmc.syky@sinopec.com
  • 基金资助:

    中国石油化工集团公司科技部项目(P15129)和国家科技重大专项(2016ZX05033-003)资助。

Multipoint geostatistical facies modeling algorithm based on pattern vector distance

Wang Mingchuan, Duan Taizhong   

  1. Sinopec Petroleum Exploration and Production Research Institute, Beijing 100083, China
  • Received:2017-10-23 Revised:2018-05-23 Online:2018-08-25 Published:2018-09-04

摘要:

多点地质统计建模算法已由基于概率模拟发展到基于相似度模拟的新阶段,基于相似度模拟的算法的关键是从训练图像中选取与数据事件最为相似的训练图型。以Simpat算法为基础发展起来的最具代表性的Filtersim和DisPat算法,采取先聚类、再匹配的建模策略,在训练图型选取的计算效率上较Simpat算法具有明显改进,但其匹配数据事件与训练图型仍采用标量距离度量二者相似度的方法。标量距离度量相似度的方法虽然具有较好的图形识别功能,但对于具有地质意义的训练图型(而非单纯的几何图形),该方法的相似度判断往往难以获取在地质含义上与数据事件最佳匹配的训练图型。考虑图型选取过程中的地质含义,兼顾实际储层的非平稳性,提出了基于图型矢量距离的多点地质统计相建模算法。新算法在训练图型和数据事件矢量距离计算的基础上,采用二次匹配方式通过矢量距离对数据事件与训练图型的相似度进行度量,最终确定与数据事件相似度最大的训练图型。以文献算例定性和定量对比了新算法和Snesim、Filtersim、DisPat算法的建模效果,并以实际油藏为例,对比了新算法与传统两点建模方法、Snesim和DisPat算法的建模效果。结果表明,新算法显著降低了训练图型选取的不确定性,所建模型更符合地质认识,从而为复杂油气藏相建模提供了一种新的方法。

关键词: 多点地质统计学, 相建模, 建模算法, 图型矢量距离, 不确定性

Abstract:

The multipoint geostatistical modeling algorithm based on probability simulation has developed into that based on similarity simulation. The key of the algorithm based on similarity simulation is to select the training pattern most similar to data event from the training image. The most representative algorithms, i.e., Filtersim and DisPat, are developed on the basis of Simpat algorithm. They adopt the modeling strategy of first clustering and then matching, thus achieve an obvious improvement in the calculation efficiency for selecting training pattern as compared with the Simpat algorithm. However, only the scalar distance is adopted to measure the similarity of data event and training pattern during matching in these algorithms. The method using scalar distance to measure similarity has a favorable pattern recognition function; however, for the training pattern with geological implication (rather than simple geometry), the similarity judgement of this method is usually difficult to obtain the most similar training pattern to data event in terms of geological implication. Considering the geological implication in the pattern selection process and the non-stationarity of actual reservoirs, a multipoint geostatistical facies modeling algorithm is proposed on a basis of pattern vector distance. According to the vector distance calculation of training pattern and data event, this new algorithm adopts twice matching, and measures the similarity between data event and training pattern through vector distance, so as to finally determine the training pattern with the most similarity to data event. Using calculation examples in literatures, qualitative and quantitative comparison analyses have been performed on the modeling effects of the new algorithm and Snesim, Filtersim and DisPat algorithms. Furthermore, taking real geological data of reservoir as an example, the comparison of modeling effects is made between the new algorithm and the traditional two-point modeling method, Snesim and DisPat algorithm. The results show that the new algorithm is able to significantly reduce the uncertainty in the training pattern selection process, and the simulation results of the new algorithm are more consistent with geological recognition. The new algorithm provides a new method for the facies modeling of complex reservoirs.

Key words: multipoint geostatistics, facies modeling, modeling algorithm, pattern vector distance, uncertainty

中图分类号: