基于神经网络的井间地震数据外推及多尺度反演

Extrapolating of the cross-well seismic data based on the neural network and multi-scale seismic joint inversion

多尺度地震资料联合反演将地面地震、井间地震和VSP等资料有机结合在一起,充分利用了不同地震资料的优点,达到了提高反演分辨率的目的。但受观测系统的限制,井间地震获得的只是一段二维剖面的信息,而地面地震是地下三维数据体的综合响应,因此无法对整个工区进行联合反演。针对这一问题,基于地层对地震波吸收的非线性系统理论,提出了利用神经网络建立地面地震数据和井间地震资料映射关系的方法。由于神经网络具有层状结构且输入、输出之间的映射关系是非线性的,从而建立地面地震数据与已有的井间地震数据之间的非线性理论模型,再将此映射关系应用到整个工区,得到高分辨率的井间地震。然后利用模型测试研究了该方法的可行性和鲁棒性。最后,将该方法应用到实际地震资料中,所得的多尺度反演结果分辨率明显提高,证明了该方法的可靠性及适应性。

The multi-scale seismic joint inversion method,which integrates the cross- well seismic data,VSP data and other seismic data with the surface seismic data organically,is fully developed the advantages of different seismic data to achieve the purpose of improving the inversion resolution.However,restricted by the observing system,the cross-well seismic data only reflects the information of a 2-D profile.The surface seismic data is the response of the three-dimensional data volume.So the joint inversion is not carried out in the whole workspace.Considering this problem,we propose a method to build the mapping relationship between the cross- well data and the surface seismic data using the neural networks based on the nonlinear system theory of the formation absorption of seismic waves.The neural network has a layered structure and the mapping relationship between the input and output is nonlinear which can describe the complex nonlinear theory model between the surface seismic data and the cross-well seismic data.Then we apply the mapping relationship to the whole workspace and obtain the cross-well seismic data with high resolution.The numerical tests show that the proposed method is feasible and noise resistent.Finally,we apply the proposed method to the field data,and the corresponding resolution of the multi-scale seismic joint inversion result is highly improved,which demonstrates the reliability and adaptability of the proposed method.