Deep learning prestack seismic inversion based on multi-seismic features

Abstract

Summary The observed pre-stack seismic data carry abundant information of underground elastic parameters, and amplitude variation with offset (AVO) is an important feature of pre-stack angle gathers. Traditional AVO inversion methods have been widely used in field data processing to achieve elastic parameter distribution. However, these methods are usually based on the linear approximation of Zoeppritz equation, and the accuracy of estimated results is open to be enhanced. In addition, the inversion stability and resolution of low-quality seismic data needs to be further improved. In recent years, deep learning method has been widely used in elastic parameter estimation, which can extract the nonlinear relationship between seismic data and elastic parameters. However, how to construct the training set is one of the crucial problems, especially when the labels are rare. The features of difference angle gathers and elastic impedance (EI) are closely related to elastic parameters, thus we propose a multi-seismic feature assisted deep learning method for elastic parameters inversion. The partially stacked angel gathers, its difference angel gathers and the corresponding EI are combined as inputs of the network. Different experimental results confirm that the proposed method has strong robustness and high inversion accuracy.