本文针对深度域地震资料反演问题提出了神经网络数据驱动岩性参数反演方法。该反演方法是非线性的,且由数据驱动,不需基于任何确定性算子,其中数据驱动方法由结构风险最小化神经网络实现。在反演运算过程中,通常将测井资料和地震记录作为训练样本,其中井旁道地震记录作为输入,测井资料作为输出,网络学习完成后,得到由地震记录转换为测井资料的反演映射关系,进而进行岩性解释。文中通过理论模型及叠前深度偏移地震资料的岩性参数反演实例,说明了方法的有效性。

In view of the issue of seismic data inversion in depth domain,the paper presented the method for inversion of lithologic parameters driven by neural network.The inversion method is non-linear and is driven by data as well as don’t need any determinate operator,among which the data-driven scheme is finished by structural-risk-minimized neural network.During the operation of inversion,the logging data and seismic data are usually taken as training samples,among which the seismic data in uphole receiver are taken as input and logging data as output,the inversion mapping relation transformed from seismic data to logging data is obtained when network learning is finished and the gained data are used for lithologic interpretation.The paper showed the effectiveness of the method by theoretic models and inversion cases of lithologic parameters of seismic prestack depth migration data.