神经网络法地震反演应用及效果

X潜山研究区属于储层非均质性极强的裂缝型油藏,构造极为复杂,初始模型的构建难度极大,多解性非常强。为此,要寻找一种方法,充分利用现有各种测井资料,弥补声波测井的不足;充分利用地震资料的中频信息,来提高储层反演的分辨率和精度,尽可能的减少反演结果的多解性,增加反演成果的客观性,以便更准确地进行储层预测。作者建立地震属性与测井资料的多维线性或非线性的统计关系,进而利用基于神经网络法地震反演和线性、非线性多元地震属性分析对裂缝型储层进行横向预测。概率神经网络方法采用了交叉检验预测精度的方法,所以,在相关性较好的情况下,增加了预测结果的可靠性,减少了多解性,提高了预测结果的精度。

Application and effect of seismic inversion based on neural network

The study area of X buried hill belongs to the fractured reservoir with very strong heterogeneity.The structure is extremely complicated,the construction of the initial model is extremely difficult and the multiple solutions are very strong.So we must make full use of logging data and middle frequency information of seismic data,make up the insufficiency of sonic logging to improve the resolution and accuracy of inversion,reduce the multiple solutions of inversion results as much as possible and increase the objectivity of inversion results so as to accurately predict reservoir.This article started from the establishment of statistics relationship between seismic attributes and log data based on linear or nonlinear relationship,and made full use of the seismic inversion based on neural network and linear and nonlinear multivariate seismic attribute analysis technology to horizontally predict the fractured reservoir.The method of probabilistic neural network prediction adopted the cross validation methods,which increased the reliability of prediction,decreased the multiple solutions and improved the prediction accuracy in the situation of good correlation.