准噶尔盆地是我国重要的含油气盆地，其勘探目标已进入深层。该盆地的复杂近地表条件、勘探目标深度以及“两宽一高”的三维地震数据采集方式导致地震资料信噪比低、数据量大，这些问题对勘探目标的落实产生了影响。因此，压制噪声并提高三维地震资料的品质对于实现勘探目标至关重要。

随着深度学习理论的发展和硬件性能的提升，深度神经网络的学习能力和处理效率得到了显著提高。为此，基于残差学习和批归一化技术，构建了三维去噪卷积神经网络(three-dimensional denoising convolutional neural network，3D-DnCNN)，并开发了适用于准噶尔盆地的基于深度学习的三维地震资料噪声压制流程。

针对准噶尔盆地某大连片工区的实际需求，选取了覆盖次数高、信噪比高的区域的噪声压制结果构建高质量标签，并将训练好的3D-DnCNN网络应用于整个工区。研究结果表明，与常规工业流程相比，所提方法得到的同相轴一致性更好、断裂保持更完整、石炭系顶界与内幕更加清晰。此外，3D-DnCNN网络在高信噪比区域学习到的偏移画弧噪声特征，使其在整个工区的偏移画弧噪声压制能力优于常规工业流程。通过调整网络参数(如网络深度、卷积核大小及训练样本选择策略)可以进一步优化网络以适应不同地区的地震资料，从而增强了地震噪声压制技术的适用性和有效性。

The Junggar Basin is recognized as a significant petroliferous basin in China, and its hydrocarbon exploration targets have shifted to deeper strata. However, the 3D seismic data of this basin suffer from low signal-to-noise ratios (SNRs) and high data volumes due to the basin's complex near-surface conditions, the great depths of exploration targets, and seismic data acquisition methods characterized by wide azimuths, broadbands, and high density. This complicates the identification of hydrocarbon exploration targets, rendering the improvement in the quality of the 3D seismic data by noise suppression vitally important.

The progress in the deep learning theory and the enhancement of hardware performance have significantly boosted the learning capability and processing efficiency of deep neural networks. Based on residual learning and batch normalization techniques, this study developed a three-dimensional denoising convolutional neural network (3D-DnCNN) and a deep learning-based noise suppression workflow applicable to the 3D seismic data of the Junggar Basin.

To meet the actual demand of a large contiguous surveyed area in the Junggar Basin, high-quality labels were constructed using the noise suppression results of zones with high seismic coverage and SNRs, and the trained 3D-DnCNN was then applied to the entire study area. Compared to the conventional industrial workflow, the workflow developed in this study yielded more consistent seismic events, more intact faults preserved, and clearer top boundary and inner layers of the Carboniferous strata. Additionally, since the 3D-DnCNN learned the characteristics of offset-related arc noise in high-SNR zones, it outperformed the conventional industrial workflow in suppressing such noise across the entire surveyed area. By adjusting network parameters such as the network depth, convolution kernel size, and the strategy for selecting training samples, the 3D-DnCNN can be further optimized to adapt to seismic data from different areas, thereby enhancing the applicability and effectiveness of the seismic noise suppression technique.