Background In seismic imaging, accurate velocity models are crucial for characterizing subsurface structures in a fine-scale manner. Notably, in coal mining, small-scale geological structures like faults and collapse columns are closely associated with mining accidents. These structures typically occur as diffracted waves in seismograms.
Objective and Methods To effectively image these small-scale geological bodies, fine-scale velocity modeling using diffracted wave information is particularly important. Hence, this study proposed a neural network-based method for analyzing diffracted-wave velocity. First, diffracted-wave velocity spectra were generated using the gathers of diffracted waves that were separated in the common virtual source domain. Second, a gamma-ray spectrum of diffracted-wave velocity ratios was generated using the quasi-linear characteristics of diffracted waves in the migrated dip-angle domain. Finally, with the conventional reflected-wave velocity spectrum and the two diffracted-wave velocity spectra as inputs, intelligent diffracted-wave velocity modeling was performed based on a convolutional attention neural network.
Results and Conclusions The tests using numerical simulation data verified that the diffracted-wave velocity analysis network can enhance the accuracy of fine-scale velocity modeling for geological bodies like stratigraphic pinch-outs and karst collapse columns. This network allows for the effective focusing of diffracted waves in the imaging profiles, thus achieving the fine-scale characterization of small- to medium-scale structures. The application of the proposed method to actual data from a coal mine further demonstrates the method's superiority over conventional methods in terms of velocity modeling efficiency and imaging accuracy. Therefore, the proposed method is more applicable to the imaging of small-scale structures under complex geologic conditions.
DownLoad: