地震特征空间到储层特征空间的非线性映射

由井资料可提供研究层段的岩相，平均孔隙率，有效厚度和砂泥岩百分比含量等多种储层参数，再上它们组成多口井的多维储层特征空间；通过对这些样本用自组织人工神经网络特征映射实现样本聚类，找出可靠的井组及每一井组所对应的储层特征类别；然后，利用井位处的地震道，通过迭代计算确定反映地震道类区分能力的特征子集，并实现对地震道的聚类；

Nonlinear mapping from seismic characteristic space to reservoir characteristic space

Many reservoir parameters of interested reservoir interval, such as lithofacies,average porosity, effective thickness and sand (or shale) percentage content ,can bederived from borehole data; they then form a characteristic space of multidimen-sional reservoir that involves several boreholes. The cluster analysis of these sam-ples is achieved by using the neural network of self-organized character mapping soas to find reliable borehole groups and the reservoir characteristic categories corre-sponding to the groups respectively' Then, the characteristic subset showing theclassification ability of seismic traces is determined by doing iterative calculation forborehole-side seismic traces; and the clustering of the seismic traces can be achievedby using the subset. Finally, a multi-layer feedforward neural network is used tomap the seismic traces to the geological space consisting of reservoir parameters soas to know lateral reservoir distribution. The article describes the example of howwe use the method to quantitatively analyse the reservoir parameter space in athree-dimensional seismic area in the northern Biyang depression.