Hybrid-Driven High-Resolution Prestack Seismic Inversion

Abstract

Pre-stack seismic inversion is considered among the most frequently utilized techniques for reservoir characterization. However, the resolution of the inverted parameters, such as P-, S-wave velocity and density, is low due to the limited band-width and side-lobe interference of the seismic wavelet. To address this issue, a hybrid two-step strategy that combines data-driven and model-driven methods is proposed to enable higher resolution and accuracy of the inverted results. We first construct a three-layer fully connected network to implement the mapping of seismic data to reflection coefficients. The method does not require exact seismic wavelet to be known and intensive human-computer interaction. It estimates reflectivity based on the extracted features from training data, which gives more accurate results compared to traditional sparse inversion methods. Then, the model-driven method (i.e., amplitude variation with offset/angle inversion method) is adopted to reconstruct P-wave velocity, S-wave velocity and density from the estimated reflection coefficients. The performance of the hybrid-driven strategy (HDS) is checked using synthetic model and real data. The results indicate that the proposed method provides more accurate and higher resolution inversion results for seismic reservoir characterization.