Seismic resolution enhancement based on GAN and its application in thin interbedded reservoir identification

【Objective】Thin interbedded reservoirs in the Triassic Akekule Formation of Tahe Oilfield are difficult to delineate using conventional seismic data because of limited bandwidth, noise and tuning effects. This study evaluates the applicability of GAN-based seismic resolution enhancement for thin interbed identification. 【Methods】Training samples constrained by real seismic noise were constructed from seismic, logging and drilling data. A GAN model with residual dense blocks and relativistic adversarial loss was used for resolution enhancement and compared with spectral shaping. The enhanced results were assessed using 25, 50 and 75 Hz Ricker-wavelet synthetic seismograms, well-seismic calibration and waveform-indication inversion. 【Results】 The dominant frequencies of the original, spectral-shaped and GAN-enhanced seismic data are approximately 24.81, 48.48 and 72.16 Hz, and their correlation coefficients with the corresponding synthetic seismograms are 0.9375, 0.8692 and 0.8347, respectively. The GAN result narrows the seismic wavelet and improves the continuity of thin-interbed boundaries and interwell anomalies in the T2a1-T2a4 intervals. The shale-content constrained inversion matches the sandstone intervals well T3 with an agreement of 81.45%.【Conclusion】 The GAN method improves the recognizability of thin-interbed seismic responses while preserving the main structural framework and wave-group relationships. However, enhanced high-frequency details should be interpreted jointly with logs, sedimentary background and drilling data.