基于混合高斯先验分布的地质统计学反演  

Geostatistical inversion based on Gaussian mixture prior distribution

作　　者：贺东阳 李海山[1] 何润 王伟 HE Dongyang;LI Haishan;HE Run;WANGWei(Research Institute of Petroleum Exploration and Development-Northwest,Lanzhou 730020,China)

机构地区：[1]中国石油勘探开发研究院西北分院,兰州730020

出　　处：《岩性油气藏》2021年第3期113-119,共7页Lithologic Reservoirs

基　　金：中国石油天然气股份有限公司勘探与生产分公司项目“消除强波阻抗地震响应对围岩反射特征影响的方法研究”(编号:kt2020-10-04)资助。

摘　　要：传统的地质统计学反演利用地质统计学模拟算法来构建模型参数的先验信息,然后在地震数据的约束下利用优化算法来获得模型参数的后验解,通常忽视了岩性对模型参数的影响并且在优化过程中计算量大。为此,将模型参数的先验分布表示为受离散岩性影响的混合高斯分布,将线性混合高斯反演理论与地质统计学的序贯模拟相结合,最后通过序贯采样的方法直接获得模型参数和离散岩性的后验解,避免了优化求解过程,且反演结果具有较好的空间连续性和稳定性。模型测试和实际资料的应用表明该方法具有较好的有效性。Traditional geostatistical inversion usually uses geostatistics simulation algorithms to construct the prior information of the model parameters,and then uses some optimization algorithms to obtain the posterior solutions of the model parameters under the constraints of seismic data,which ignores the influence of lithology on model parameters and requires a lot of calculation in the optimization process.Therefore,the prior distribution of model parameters was expressed as Gaussian mixture distribution influenced by discrete lithology,then the linear Gaussian mixture inversion theory was combined with sequential simulation of geostatistics,and the posterior solutions of the model parameters and discrete lithology were directly obtained by sequential sampling,which avoids the optimization process,and the inversion results have good spatial continuity and stability.Both model tests and the application of actual data show the effectiveness of the method.

关 键 词：地质统计学 序贯模拟 混合高斯 离散岩性 

分 类 号：P631.4[天文地球—地质矿产勘探]