Improving the resolution of poststack seismic data based on UNet+GRU deep learning method

Abstract

Most existing seismic data frequency enhancement methods have limitations. Given the advantages and disadvantages of these methods, this study attempts to apply deep learning technology to improve seismic data resolution. First, on the basis of the UNet deep learning method, which combines well-logging and seismic data, a synthetic seismic record is established with logging acoustic data and density, the borehole synthetic seismic record is labeled, and the borehole seismic trace data are taken as the input data. The training model of the borehole seismic trace data and the borehole synthetic seismic record is established to improve the medium- and high-frequency information in the seismic data. Second, the gate recurrent unit (GRU) is used to retain the low-frequency trend in the original seismic record, and the UNet and GRU results are combined to improve the medium- and high-frequency information while preserving the low-frequency information in the seismic data. Then, model training is performed, the model is applied to the three-dimensional seismic data volume for calculation, and the seismic data resolution is improved. The information extracted using our method is more abundant than that extracted using previous methods. The application of a theoretical model and actual field data shows that our method is effective in improving the resolution of poststack seismic data.