Efficient blind seismic impedance inversion based on a deep learning accelerated alternative iteration inversion method

We propose the use of deep learning to improve the efficiency of blind seismic impedance inversion. The method consists of the following three steps: (1) Extraction of a small number of 2D profiles from 3D seismic data and conduct blind seismic inversion with a traditional method to obtain the AI model and wavelet of these profiles; (2) Using the inversion results of the first step as label data, the deep neural network is trained to learn the nonlinear mapping of post-stack seismic data to AI and wavelet; (3) The depth network is used to predict the AI and wavelet of most other 2D seismic profiles, and the regularization terms are constructed based on the prediction results to constrain the alternate iterative inversion of the remaining 2D profiles. Because of the high accuracy of the network prediction results, the new alternate iterative inversion method converges quickly and the terms are easy to select. The experimental results based on synthetic and field data examples verify that the proposed method has significant advantages over the traditional method in terms of efficiency and inversion accuracy.