CNN-GRU模型在克里金插值中的应用  

Application of CNN-GRU model in Kriging interpolation

作　　者：郭天良 宋强功 郭淑文[3] 许辉群[1] GUO Tianliang;SONG Qianggong;GUO Shuwen;XU Huiqun(College of Geophysics and Petroleum Resources,Yangtze University,Wuhan,Hubei 430100,China;BPG Inc.,CNPC,Zhuozhou,Hebei 072751,China;Research Institute of Exploration and Development in Dagang Oilfield,PetroChina,Tianjin 300280,China)

机构地区：[1]长江大学地球物理与石油资源学院,湖北武汉430100 [2]东方地球物理公司,河北涿州072751 [3]中国石油大港油田勘探开发研究院,天津300280

出　　处：《石油地球物理勘探》2025年第1期185-192,共8页Oil Geophysical Prospecting

基　　金：中国石油基础研究专项“多物理场融合及储层成像方法研究”(2023ZZ05-05)资助。

摘　　要：克里金插值是一种可以结合经验知识的建模方法,其中变差函数的求取精度决定了插值的效果,从而影响基于克里金插值的地震反演低频模型的构建。传统的克里金插值方法难以同时使用多个不同的变差函数理论模型来提高低频模型构建的精度,而仅仅利用单一的理论模型实现变差函数求解,存在理论模型选择的不确定性、变差函数拟合值偏低的平滑效应以及井距较远产生的空洞效应。为此,引入神经网络CNN-GRU模型,能够自适应拟合向量到对应井之间半方差的复杂关系,进一步实现球状模型、高斯模型、指数模型和空洞效应模型的有效融合,从而解决变差函数的不确定性、平滑效应和空洞效应。该模型考虑了井间的相关性,可便捷地实现逐点的变差分析,处理过程方便,可较好匹配变差函数选取参数的随机性。实际资料应用表明,基于CNN-GRU模型的克里金法可建立一个高精度的低频地震反演模型,其效果相较于传统方法更优。Kriging interpolation is a modeling method that can be combined with empirical knowledge,in which the accuracy of variogram determines the effect of the interpolation,thus affecting the construction of low-frequency model of seismic inversion based on Kriging interpolation.It is difficult for the traditional Kriging interpolation method to use multiple different theoretical models of variogram at the same time to improve the accuracy of low-frequency model construction,and only using a single theoretical model to solve the variogram leads to the uncertainty of theoretical model selection,the smoothing effect with a low fitting value of the variogram,and the hole effect caused by a long well distance.To solve the above problems,a neural network CNN-GRU model is introduced to adaptively fit the complex semi-variance relationship between the vector and the corresponding well and further realize the effective fusion of the spherical model,the Gaussian model,the exponential model,and the hole effect model,so as to address the uncertainty,the smoothing effect,and the hole effect of the variogram.The model takes into account the correlation between wells and conveniently realizes the point-by-point variation analysis with a convenient processing process,which can well match the randomness of the parameters selected by the variogram.The actual data show that the Kriging method based on the CNN-GRU model can establish a high-precision low-frequency model of seismic inversion,and it has a better effect than the traditional method.

关 键 词：CNN-GRU 变差函数 克里金 低频模型 

分 类 号：P631[天文地球—地质矿产勘探]