Building Low-frequency model for reservoir characterization based on deep learning

Inspired by domain seismic inversion method in the complex frequency domain and sequence modeling network, we propose a semi-supervised spatial sequence method in the complex frequency domain to build initial low-frequency model of reservoir parameters. The proposed method considers spatial information to ensure the horizontal continuity of the estimated low-frequency model. Based on the advantage that the low-frequency information of seismic signals in the complex frequency domain is easier to obtain, we use the forward model in the complex frequency domain to replace the time forward model. The proposed workflow was validated on the Marmousi II model. Although the training was carried out on a small number of low-frequency impedance, the proposed workflow was able to build low-frequency model for the entire Marmousi II model with an average correlation of 98%. Taking the model obtained by the proposed method as the initial low-frequency model of the conventional inversion method, better impedance results can be estimated.