地震信息的属性参数提取和砂体预测方法

Method for making both seismic attributive parameter extraction and sand body prediction

在利用地震信息属性参数进行砂体预测时，仅用单一参数预测的结果往往精度很差，而盲目使用多参数作为神经网络的输入，又会使网络的学习过程不收敛。为克服上述问题，本文通过理论模型研究，并结合实际地震资料，从时间域和频率域中提取了目的层的19个地震信息属性参数。然后，选取与薄砂层厚度最密切的8种参数进行砂体预测。文中对几种常用的预测方法进行了分析和对比。应用结果表明，多参数的神经网络预测方法的精度最高；主频线性预测和振幅频率反演预测方法次之；振幅线性预测方法的精度最差。另外，当目的层最大砂岩厚度超过λ／4时，用频率域参数预测砂岩厚度比用时间域参数预测的精度高。

Single parameter prediction usually brings poor accuracy in the sand body prediction that is achieved by using seismic attributive parameters. Nevertheless, blindusing the multiple parameters as the input of neural network will make nonconvergent the learning course of the neural network.After theoretical model research,19 seismic attributive parameters in time and frequency domains can be derived fromthe seismic informations of objective interval so as to cope with above problems.Then 8 parameters which are very related to sand body thickness are used to predictsand bodies. In addition, the usual prediction methods are analysed and comparedin detail.The trial results draw the conclusion: multiparameter neural network prediction brings satisfactory accuracy, dominant-frequency linear predicti0n and amplitude-frequency inversion prediction give less satisfactory accuracy, and amplitudelinear prediction causes poor accuracy. When the objective interval is thicker than/4, sand thickness prediction using frequency-domain parameters is more accuratethan sand thickness prediction that introduces time-domain parameters.