BP神经网络与多点地质统计相结合的井震约束浊积水道模拟

Well-seismic Data Constrained Turbidite Channel Modeling Based on the Combination of Neural Network BP Arithmetic and Multiple-point Geostatistics

勘探阶段,仅依靠数目较少的井点数据很难在较大工区范围内建立令人信服的沉积相模式。然而,以地震数据做约束,往往会使沉积相更加逼近软数据的分布趋势。以澳西北陆架Vulcan次盆P22区块Puffin组Unit7小层为研究对象,在建立沉积相的过程中,提出了分层次约束的方法,即以BP神经网络算法控宏观,以多点地质统计学约微观的方法流程,同时结合已知井的岩性类型数据点和源于地震的岩性类型数据点,再现了浊积水道的沉积展布规律。采用截断高斯和多点地质统计学2种方法进行模拟,分别产生了浊积水道的8个模拟实现,对比结果表明,多点地质统计学模拟方法在再现浊积水道的几何形态方面比基于变差函数的截断高斯模拟更具有优越性。最后以偏差目标函数为标准,优选出了偏差最小的模拟实现。

At oil exploriation stage,it was hard to build the convinced sedimentary facies modes with less well-point data in the large operation area.However,for the restriction of seismic data,the sedimentary facies was often more approximate to the distributive tendency of soft data.In the process of building the sedimentary facies,Subzone 7 of Puffin Formation in Block P22 in the Vulcan Sub-basin of north-west Australia continent shelf was used as studied object,a layering constraining method was proposed.The idea that macroscale of modeling was controlled by BP arithmetic and micro-shape of geological body was depicted by multiple-point geostatistics propounded in the modeling progress,and in combination with the lithologic type data derived from known well data and seismic data,the distribution of turbidite channel could be realized by modeling.Truncated Gaussian simulation and multiple-point geostatistics are applied in the modeling process,and each method responds to 8 simulation processes respectively.The comparison result indicates that multiple-point geostatistics is more proper than variogram based Truncated Gaussian simulation in the process of characterizing the geometry of turbidite channel.The optimal simulation with less errors is obtained by the criterion of minimal value of bias object function.